Family digital genealogy system based on double-track data processing and construction method

Abstract

The application belongs to the technical field of digital humanities, and discloses a family digital genealogy system and a construction method based on double-track data processing. The system is deployed in a home local area network edge computing terminal which is physically isolated from the external public Internet. The system comprises a multi-source data access layer, a double-track data processing engine, a three-dimensional digital genealogy modeling engine, a local area network interaction module and a privacy destruction module. Through parallel generative backtracking and original time sequence tracks, the double-track data is complemented by using a globally unified identity. The system constructs an attribute knowledge graph locally, and supports large model offline interaction in the local area network. In asset delivery and management, hardware-level data erasure is provided in the double scenarios of service provider field equipment delivery and destruction and user local terminal self-destruction, and the hash value of the destruction report is synchronized offline to the alliance chain for storage. The application guarantees the local area network absolute data sovereignty, full life cycle inheritance and compliant verifiable destruction of family private data.

Description

Technical Field

[0001] This invention belongs to the fields of digital humanities, edge computing and artificial intelligence privacy and security technology, and specifically relates to a family digital genealogy system and construction method based on dual-track data processing. Background Technology

[0002] With the widespread adoption of generative artificial intelligence, home edge computing devices, and wearable IoT, the demand for data preservation of an individual's entire life cycle and the digital transmission of family and cultural records continues to surge. Currently, the market primarily offers traditional electronic family trees, cloud-based AI digital humans, and smart health wearable devices as the main solutions. However, in-depth analysis of existing technologies reveals the following intractable structural flaws in the industry:

[0003] First, centralized cloud storage completely deprives users of their data sovereignty, posing extremely high risks to privacy and security. Current systems heavily rely on public cloud storage for sensitive data such as user biometrics, private family videos, and recorded conversations. Due to the openness of the public internet, the highly private data of high-net-worth families faces significant risks, including unauthorized access by platforms, leaks due to hacker attacks, data loss due to service provider outages, and even unauthorized use for training third-party large-scale models. Existing data privacy protection urgently needs fundamental innovation towards an "edge-side, locally isolated" model.

[0004] Second, there is a fragmented and disjointed data structure across the entire life cycle. Individual data is scattered across heterogeneous terminals with incompatible interfaces. Traditional genealogies only statically record structured household registration information, and smart wearables only record physiological health data for short periods, failing to create a continuous life record from birth, growth, middle age to old age, resulting in serious blind spots in time coverage.

[0005] Third, the data processing model is singular, making it difficult to achieve bidirectional complementary integration of heterogeneous data. Existing technologies only support a single route: post-event subjective verbal recollection or online objective sensor data acquisition. Both lack a unified hardware platform and standardized mapping protocol. Objective raw data cannot correct memory biases in subjective verbal recollections, and subjective verbal recollections cannot imbue objective data with a spiritual core, leaving heterogeneous data in a state of perpetual fragmentation.

[0006] Fourth, genealogy modeling is limited to a single dimension and lacks the ability to digitize implicit spiritual assets. Traditional and existing electronic genealogies are mostly limited to explicit bloodline information such as birth and death dates and lineage. They cannot perform in-depth structured modeling of high-value implicit spiritual assets such as an individual's family emotional choices, the logic of overcoming life's darkest moments, the cause and effect of career decisions, and family traditions and anecdotes. They also lack the ability to deeply and intelligently transmit information across generations.

[0007] Fifth, there is a lack of compliant and verifiable evidence for data destruction, resulting in significant vulnerabilities in privacy compliance. Existing products primarily employ backend logical deletion (i.e., modifying data pointers) rather than achieving underlying overwriting at the physical media level. Particularly in commercial procurement scenarios, temporary sensitive data on service providers' field devices cannot be supported by legally valid destruction certificates, and third parties cannot verify the authenticity of the deletion, making it difficult to meet the stringent compliance audit requirements of laws such as the Data Security Law. Summary of the Invention

[0008] This invention aims to at least partially solve the aforementioned technical problems. Therefore, the purpose of this invention is to provide a family digital genealogy system and its construction method based on dual-track data processing. This invention locks the entire business process within a completely physically isolated local area network environment, innovatively employing a dual-track heterogeneous data complementary architecture, three-dimensional mental graph modeling, and dual-scenario hardware-level compliant destruction, thus completely solving the technical bottlenecks of digital genealogy in terms of privacy sovereignty, full-cycle coverage, and compliant destruction.

[0009] The technical solution adopted in this invention is as follows:

[0010] A family digital genealogy system based on dual-track data processing is deployed in a local private AI computing terminal that is completely physically isolated from external public networks. It supports independent offline operation within a home LAN environment, and the original multimodal data is strictly prohibited from being transmitted to the public network throughout the entire process. The system includes:

[0011] The multi-source data acquisition interface layer is used to access and acquire multimodal raw data of family members at different life stages within the home LAN. The opening of the IoT hardware acquisition channel requires separate authorization from the user's liveness.

[0012] The dual-track data processing engine, connected to the multi-source data acquisition interface layer, includes a parallel generative backtracking processing track and a native time-series recording track. It is used to divert the multimodal raw data to the corresponding track for differentiated processing based on the age characteristics of family members, and to achieve the same-source fusion of dual-track data through a globally unified kinship identification.

[0013] The digital genealogy modeling engine, connected to the dual-track data processing engine, is used to construct multi-dimensional graphs from the fused data, generating a structured digital genealogy model that includes emotional, career, and inheritance dimensions, and storing it in a local offline graph database.

[0014] The interaction module is used to receive natural language query commands from users within the local area network, route the commands to the corresponding graph dimension based on intent recognition, and drive the local large language model to generate enhanced search answers.

[0015] The intergenerational inheritance module is used to configure hierarchical access permissions within the family LAN, and automatically freezes the modification permissions of the corresponding data after verifying the user's death status, switching to a read-only digital ancestor interaction mode.

[0016] The privacy protection and data destruction module is used to perform hardware-level encrypted storage of all data in the terminal based on national cryptographic algorithms. The module supports dual execution scenarios of "service provider field equipment delivery destruction" and "user local terminal self-destruction". When destruction is triggered, physical-level data erasure based on national and international standards is performed, and the destruction report is written to a one-time read-only log area. Its hash digest can be synchronized to the consortium blockchain for evidence storage through offline physical media.

[0017] Furthermore, the local private AI computing terminal is a hardware device that integrates edge heterogeneous computing power units, and its motherboard has a hard-wired hardware encryption chip that supports national cryptographic algorithms; the hardware encryption chip has a built-in true random number generator, which is used to generate, derive and manage SM4-GCM data encryption keys in the secure area inside the chip, and the keys are physically prohibited from being exported to the outside of the chip.

[0018] Furthermore, the diversion and processing logic of the dual-track data processing engine specifically includes: for users whose age is greater than or equal to a preset advanced age threshold, or who lack continuous native data, the data is diverted to the generative backtracking processing track. A locally deployed lightweight AI model is used to complete and structure the existing historical unstructured materials, and implicit watermarks and explicit interface identifiers are forcibly embedded at the bottom layer of the AI-generated content; for natural persons whose age is less than the preset advanced age threshold, the data is diverted to the native time-series recording track. The IoT collection channels in the local area network are dynamically started and stopped according to the preset life cycle age range. The multimodal raw time-series data is stored with lossless encryption and periodically timestamped indexes are generated, prohibiting AI semantic rewriting and content extraction; when a user in the native time-series recording track reaches the preset advanced age threshold, their objective time-series behavioral data is automatically desensitized and extracted, and used as supplementary material for cross-track unidirectional input to the generative backtracking processing track.

[0019] Furthermore, the digital genealogy modeling engine adopts an attribute graph data model, specifically including three orthogonal sub-models: the emotional relationship sub-model, built on a graph attention network, used to represent the strength of emotional relationships among family members and the process of individual emotional choices; the event-related sub-model, built on a temporal graph neural network, used to represent the causal logic, success and failure attribution, and event chain of family members' career decisions; and the inheritance sub-model, used to construct a temporal graph representing the family's geographical migration and the iteration of family traditions and culture. All three orthogonal sub-models achieve cross-dimensional semantic association through the globally unified kinship identity identifier.

[0020] Furthermore, the interaction module incorporates a lightweight intent classification model based on a deep twin network architecture, used to segment user semantics in real time within a local area network; it adopts a GraphRAG retrieval-enhanced generation architecture to retrieve original factual materials from the three-dimensional graph, driving a local offline large model to generate answers that fit the logic of the parties involved, and forcibly attaching AI-generated risk warning information at the end of the response.

[0021] Furthermore, the intergenerational transfer module is configured with four levels of differentiated intranet access permissions: read-only question and answer, image viewing, data export, and model editing; and supports full offline export of digital assets through encrypted physical media when the network is disconnected.

[0022] Furthermore, in the dual execution scenario, the privacy protection and data destruction module executes a unified hardware-level erasure standard: for magnetic storage media, it continuously performs three full-disk overwrite operations, including full overwrite of 0x00, full overwrite of 0xFF, and true random numbers; for semiconductor storage media, it issues an ATA underlying security erasure command and simultaneously physically erases the data encryption key inside the hardware encryption chip; after the destruction is completed, a destruction report containing the device's unique identification code, operation timestamp, and liveness verification record is generated and permanently stored as a local, tamper-proof credential.

[0023] This invention also provides a method for constructing a family digital genealogy based on dual-track data processing, applied to the aforementioned system. The method is executed entirely in a local area network environment physically isolated from public networks, and includes the following steps:

[0024] Step S1: Deploy a local private AI computing terminal in the home LAN environment, initialize the identity information of family members and generate a globally unified kinship identity mapping table, and synchronize the hash digest of the initialized data to the consortium blockchain for basic evidence storage through offline encrypted media;

[0025] Step S2: Obtain the target user's age information and local native data storage, and based on the dual-dimensional judgment logic, divert the user to the generative backtracking processing track or the native time-series recording track;

[0026] Step S3: In the generative backtracking process, based on the existing unstructured materials, use the local offline large model to perform structured sorting and lightweight AI completion, and embed generative watermarks;

[0027] Step S4: In the native time-series recording track, the local area network dynamically activates the corresponding IoT acquisition channel, performs national cryptographic hardware-level encryption storage on the raw sensor data, and periodically closes to generate an annual memory index; when the track user reaches the preset old age threshold, their time-series data is cross-tracked to supplement the generative backtracking processing track.

[0028] Step S5: Based on the globally unified kinship identity identifier, the dual-track heterogeneous data are fused together to construct a digital genealogy model in this map database that includes three-dimensional sub-maps of emotion, career, and inheritance.

[0029] Step S6: Provide natural language human-computer interaction services within the local area network based on a preset hierarchical authorization mechanism, and automatically switch the account to read-only digital ancestor mode after verifying the user's death;

[0030] Step S7: When the service provider completes the delivery of field materials or when the user actively initiates a clearing command, the corresponding hardware-level data destruction process is triggered, a destruction report containing a timestamp is generated and written to the local read-only log area, and the hash digest is supported to be synchronized offline to the blockchain for third-party verification.

[0031] Furthermore, in the step of triggering the corresponding hardware-level data destruction process: if it is a service provider's field material delivery scenario, then after the data from the field collection device is completely imported into the home LAN terminal, the hardware-level destruction process is immediately executed on the field device to clear the temporarily collected sensitive data; if it is a user-initiated clearing scenario, then within the home LAN terminal, after authentication, the hardware-level destruction process is executed on the specified local storage medium.

[0032] Furthermore, the hardware-level data destruction process specifically includes: performing underlying security erasure and physical erasure of encryption keys on semiconductor storage media; performing multiple data overwrites on magnetic storage media; and finally writing the generated compliant destruction report into the one-time read-only log area (OTP) to ensure that the destruction behavior is irreversible and traceable at the physical level.

[0033] The beneficial effects of this invention are as follows:

[0034] 1. The system of this invention is entirely deployed on a home intranet terminal, physically isolated from the public internet, with zero cloud transmission of raw multimodal data. It is equipped with a national level-two cryptographic hardware encryption chip, and the data encryption key is truly randomly generated within the chip and never exported. From the three dimensions of hardware carrier, cryptographic algorithm, and network transmission, it creates an absolutely secure digital asset haven for high-net-worth families, completely eliminating the risk of cloud leakage.

[0035] 2. Addressing the pain point of fragmented lifecycle data, this invention designs a rigid, bidirectional closed-loop track. The generative retrospective track extracts subjective spiritual assets for the elderly, while the native time-series track preserves objective, intact life traces for younger individuals. More importantly, as users on the time-series track age, their accumulated objective data unidirectionally crosses tracks to correct subjective narrative biases, achieving a complementary and integrated fusion of heterogeneous data across the entire lifecycle, from newborns to deceased elders.

[0036] 3. This invention breaks through the limitations of traditional genealogies that only record explicit bloodlines. It designs a three-dimensional orthogonal modeling framework encompassing emotional, practical, and inheritance aspects, and utilizes cutting-edge graph neural network models such as GAT and TGN to deeply transform fragmented experiences of family elders, such as emotional choices, career breakthroughs, and family wisdom, into a structured and searchable digital genealogy, thus endowing the digital genealogy with extremely high humanistic value.

[0037] 4. This invention precisely addresses the pain points of commercial procurement and self-management, clearly distinguishing between two independent scenarios: the delivery and destruction of field equipment and the self-destruction of terminals. It abandons invalid logical deletion and enforces dual physical erasure standards (GB 46864-2025 and DOD 5220.22-M) (ATA underlying erasure, physical circuit breaker key, and full disk overwrite). Through local OTP hardware log retention and offline media hashing on-chain architecture, it achieves a highly verifiable audit loop while maintaining network physical isolation.

[0038] 5. In a network-disconnected environment, the system of this invention independently runs an INT8-quantized fine-tuned large language model and a retrieval enhancement architecture (GraphRAG) using the heterogeneous computing power of the NPU / GPU on local edge devices. The AI-generated interactive responses completely strip away the linguistic templates of general large models, precisely matching the original thought processes of our predecessors. This feature is perfectly suited for high-end private scenarios such as overseas retreats and villas without external network access, achieving truly sustainable inheritance unrestricted by external networks and the lifespan of commercial cloud services. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the architecture of the family digital genealogy system based on dual-track data processing according to the present invention;

[0040] Figure 2 This is a schematic diagram of the process for determining, diverting, and integrating the dual-track technology route in this invention.

[0041] Figure 3 This is a schematic diagram of the hardware-level data destruction and blockchain evidence verification process under dual scenarios of the present invention;

[0042] Figure 4 This is a flowchart illustrating the steps of the family digital genealogy construction method based on dual-track data processing according to the present invention. Detailed Implementation

[0043] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0044] like Figure 1 As shown, this embodiment provides a family digital genealogy system based on dual-track data processing. The system is hosted in a customized local private AI computing terminal. In terms of physical hardware, the terminal can be deployed as an edge computing NAS host in a home's low-voltage electrical box or study. The terminal motherboard is equipped with a CPU (Central Processing Unit) and a GPU / NPU (Neural Processing Unit) to provide local heterogeneous computing power of no less than 20 TOPS, and is configured with a large-capacity solid-state drive (SSD) and hard disk drive (HDD) array (supporting RAID 1 redundancy).

[0045] To achieve data isolation, the terminal implements physical restrictions at the network interface level. The system kernel directly blocks external wide area network (WAN) routing at the underlying level, only opening local area network (LAN) communication based on the IEEE 802.11 standard and Ethernet protocol, so that it can run even when the network is disconnected.

[0046] Regarding encryption hardware, the terminal motherboard integrates a national level 2 encryption chip (such as the National Technology N32 series security chip) certified by the State Commercial Cryptography Administration. This chip incorporates a True Random Number Generator (TRNG) conforming to the GM / T 0005-2021 standard. During system initialization, the TRNG generates a master key within the chip's secure isolation area and derives an SM4-GCM working key for full-disk data encryption. This working key is physically and logically permanently sealed within the chip's secure area; even with root privileges, the external operating system cannot export it. Any physical disassembly or copying of the hard drive will only retrieve high-strength garbled encrypted text.

[0047] The system is equipped with a multi-source data acquisition interface layer, supporting access to IoT devices such as cameras and smart bracelets within a home LAN via proprietary protocols such as RTSP / MQTT. It also supports offline import of scanned copies of paper documents and old videos via encrypted USB flash drives. Specifically, for any routine IoT data acquisition channel, the system requires users to perform a single liveness detection authorization via the terminal camera within the LAN. The maximum validity period for a single authorization is set at one year; after expiration, the underlying channel will be immediately shut down, absolutely eliminating the risk of silent eavesdropping or unauthorized recording.

[0048] refer to Figure 2 The collected heterogeneous data enters the dual-track data processing engine. The system generates a globally unified kinship identifier (UUID) for each registered family member, serving as a unique primary key across business lines and data graphs. The engine has a built-in two-dimensional decision-making and routing unit, performing the following rigid routing based on age characteristics and data volume:

[0049] 1. Generative Retrospective Processing Track: This track is designed for living seniors aged 60 or older (preferably 60 years old in this embodiment) or deceased ancestors without continuous digital traces. This track primarily processes unstructured historical materials such as old photos and oral recordings. The system utilizes local NPU computing power, loads local large language models such as MiniCPM-2B (quantized and compressed with INT8), and combines LoRA (low-rank adaptation) fine-tuning technology to automatically organize the disorganized oral recordings into structured slices of life events. For missing images of ancestors, the system activates specific timbre parameters through a local hybrid expert model (MoE architecture) to perform high-fidelity voice cloning and calls a lightweight LivePortrait facial expression transfer model to generate a dynamic digital human. To prevent AI from tampering with history, all AI-completed content is forcibly embedded with an invisible global implicit watermark in the underlying pixel matrix or audio frequency domain, and is prominently displayed in a prominent position on the user interface.

[0050] 2. Native Time-Series Recording Track: Designed for individuals under 60 years of age and newborns. This track adheres to the principle of absolutely objective recording, eliminating subjective intervention. For different life stages such as 0-3 years, 3-12 years, and 12-18 years, the system dynamically matches local area network data collection channels including infant monitoring mattresses, children's smart wearables, and home environment sensors. The collected multimodal data (such as sleep heart rate curves, movement trajectories, and native home videos) is directly encrypted with SM4 using the aforementioned national cryptographic chip and stored sequentially in the database. The system only periodically (e.g., at the end of each year) extracts timestamps using a sliding time window algorithm to generate an annual memory index, prohibiting AI large models from semantically rewriting or emotionally labeled data extraction throughout the process.

[0051] 3. Dual-track cross-track integration: The system's breakthrough lies in the fact that when younger users in the native time-series track reach the 60-year age threshold, the engine automatically extracts their decades-long accumulated objective health data curves and behavioral pattern changes. After desensitizing sensitive privacy fields, this native time-series data is fed back unidirectionally into the generative retrospective processing track, serving as an objective anchor point for their oral recollections and accurately correcting memory biases and subjective distortions caused by aging.

[0052] The merged data is fed into the digital genealogy modeling engine. This system abandons traditional relational databases, deploying a lightweight Neo4j graph database locally and employing a property graph data model. It utilizes graph neural network (GNN) technology to construct a highly humanistic and orthogonal three-dimensional digital genealogy.

[0053] The Emotional Connection Model focuses on an individual's inner emotions and intimate relationships. Based on a Graph Attention Network (GAT), it models the allocation of emotional weights in the memories of the elderly. Nodes represent family members, and the weights of edges represent the strength of emotional dependence. It also records key emotional decision-making events (such as marriage and separation).

[0054] The "Important Sub-Model" focuses on life vision and decision-making wisdom. Based on the Temporal Graph Neural Network (TGN), it extracts an individual's career ups and downs, major asset investments, and crisis response measures into a chain of events with causal logic, enabling future generations to directly retrieve the success and failure attribution patterns of their ancestors when facing similar predicaments.

[0055] The Inheritance and Foundation Model focuses on the macro-level family foundation. It constructs a temporally changing geographical migration map and the iterative process of family traditions and precepts, forming the basis for a grand family narrative. These three sub-models all rely on relatives' UUIDs for efficient cross-graphological connections, constructing a panoramic and three-dimensional personality knowledge graph.

[0056] In terms of interactive experience, the interaction module receives natural language questions input by descendants within the local area network (e.g., "How did your grandfather adjust his mindset when he encountered a career slump in 1998?"). The system first utilizes an intent classification model based on a Deep Siamese Network architecture (inference latency less than 50ms) to accurately split the question intent and route it to a composite dimension of "matter-related" and "emotional." Subsequently, the system adopts a local area network retrieval enhanced generation architecture (GraphRAG), skipping traditional text matching and directly retrieving relevant entity nodes and subgraph paths from the local graph database. Based on this, the local large language model generates an answer that is full of humanistic care and conforms to the logic of ancestors, and forcibly adds a compliance prompt at the end of the answer: "AI generated based on historical inference, for reference only."

[0057] The intergenerational inheritance module grants family administrators powerful intranet access control capabilities. It can be finely configured with four levels of permissions: "read-only Q&A" (limited to intranet inquiries only), "image viewing" (allowing access to original videos), "data export" (supporting offline copying), and "model editing" (allowing modification of materials). When the system verifies the death of a family member through an anonymized data interface, a linked logic is triggered: the underlying database locks write permissions for that member's record, irreversibly switching their account status to a perpetual "read-only digital ancestor" mode.

[0058] refer to Figure 3 The privacy protection and data destruction module is a core innovation of this system. In real-world business operations, there are inevitably scenarios where service teams collect data on-site and where users themselves want to erase their records locally. To address these issues, this system has designed two independently triggered but standardized hardware-level destruction processes.

[0059] Scenario 1: Delivery and destruction of service provider's field equipment. This refers to the field data collection kits carried by service providers. Because these kits collect a large amount of original audio and private video footage of elderly people during field interviews, once this temporary sensitive data is completely and encryptedly imported into the user's local private AI computing terminal via the local area network, the system forcibly triggers the data destruction of the field equipment.

[0060] Scenario 2: User's local terminal self-destruction. After completing liveness authentication on the intranet, family users proactively initiate a clear command to delete the data stored locally.

[0061] In both scenarios, the system refuses to use simple logical pointer modification operations, and instead executes physical-level erase commands that comply with both the GB46864-2025 national standard and the US Department of Defense DOD 5220.22-M international standard.

[0062] For semiconductor storage media (such as SSDs in field acquisition boxes or terminal system disks), the ATA Secure Erase low-level command is directly sent to the main control chip to instantly clear the physical blocks of the NAND flash memory under high voltage; at the same time, cryptographic erasure is performed, that is, a destruction command is sent to the encryption chip to physically melt the SM4 encryption key sealed inside, making the entire disk of ciphertext instantly become unbreakable digital gibberish.

[0063] For magnetic storage media (such as HDDs used for large-capacity backups in terminals), the underlying driver ignores the file system and performs three physical sector overwrites in succession: the first writes 0x00 to the entire disk, the second writes 0xFF to the entire disk, and the third overwrites the entire disk with a true random number, and performs sector verification after each overwrite.

[0064] After the destruction process is completed, the terminal system generates a highly detailed destruction report, including the device's unique serial number (SN), millisecond-level timestamps, the operator's liveness verification hash value, and the specific overwrite algorithm used. This report is hard-written into the OTP (One-Time Programmable Read-Only Memory) log area on the motherboard, achieving physical immutability. To form a perfect compliance audit loop, the report's hash digest can be exported via an encrypted offline USB drive and uploaded to an offline PBFT consensus consortium blockchain node for evidence storage via a one-way synchronization gateway. Third-party regulatory agencies can issue a legally valid data erasure compliance certificate by comparing the on-chain hash with the local OTP log and using magnetic microscopy for further examination.

[0065] refer to Figure 4 The method flow shown below will be illustrated in detail through several specific embodiments to demonstrate the operating mechanism of this system.

[0066] Example 1: Generative Retrospective Trajectory and Field Destruction Scenario for Living Elderly Individuals

[0067] There was a 78-year-old member of the family whose life story was rich but lacked electronic records. The service team went to his residence with disconnected internet access to conduct a week-long closed-door interview, collecting a large amount of audio and paper biographical manuscripts.

[0068] After the field data collection is completed, the service team connects the field device to the elderly person's home intranet (step S1). The system assigns a unique UUID mapping for the elderly person within the local area network, and then imports the temporary data dictated by the elderly person into the local computing terminal. After the data integrity verification is passed, the field device immediately triggers scenario one (field delivery and destruction), performing ATA-based security erasure and key circuit breaking on its SSD. Subsequently, the system writes the destruction hash digest to a USB flash drive, which the service team brings back offline to the blockchain, ensuring that not a single byte of the customer's actual data is taken away.

[0069] In the local home terminal, the elderly person is identified as being in the generative backtracking process due to their age (>60 years old) (step S2). The MiniCPM large model inside the terminal performs semantic segmentation on the biographical manuscript offline (step S3), extracts the causal logic of their early entrepreneurship in Southeast Asia, and constructs a "related" graph in the local Neo4j database (step S5). Three years later, the elderly person passes away, and the system automatically locks their status to read-only ancestor mode (step S6). Their grandchildren, in a study without external network access, ask the screen via the local area network, "Grandpa, how should we break the deadlock when we encountered vicious competition from our peers?" The system, based on GraphRAG, retrieves the elderly overseas Chinese's real decision-making graph from that time and generates a wise answer that fully reflects their original will.

[0070] Example 2: Native Recording and Homologous Fusion of the Entire Life Cycle of Newborns

[0071] A newborn is born, and the family administrator initializes and stores its identity on the intranet (step S1). Due to its age (<60 years old), it is seamlessly diverted to the native time-series recording track by the system (step S2).

[0072] During the 0-3 year period, the system dynamically activates the smart mattress and camera within the baby's room LAN; as the baby grows to 15 years old, the system automatically switches to the authorized LAN channel compatible with their smart fitness tracker; at age 30, the system collects the trajectory of their smart home and in-vehicle systems. All of this time-series data completely avoids excessive intervention from AI semantics, and is stored only through SM4 hardware encryption, and is periodically closed into an annual memory timestamp index (step S4).

[0073] As time passed, the individual celebrated his 60th birthday. Upon crossing the preset age threshold, the system triggered a cross-track mechanism: it extracted and anonymized the massive amounts of raw data accumulated over the past 60 years, including health change curves and long-term geographical migration heat maps, and injected this solid foundational evidence into his newly begun oral memoir, recorded in a generative track. If his narration contained age-related timeline inconsistencies, the system could use objective data to correct these inconsistencies, completing a closed-loop data fusion spanning sixty years (step S5). Subsequently, even if the family triggered local data self-destruction under special circumstances (step S7), its robust execution chain and compliance logs could still withstand rigorous legal scrutiny.

[0074] The above description is merely a specific embodiment of the present invention under the core design concepts of local area network physical isolation, dual-track fusion, and hardware-level compliant destruction. Any adaptive adjustments and equivalent substitutions made by those skilled in the art to computing power allocation, model selection, and network communication protocols without departing from the core principles of the present invention should be covered within the patent protection scope of the present invention.

Claims

1. A family digital genealogy system based on dual-track data processing, characterized in that, The system is deployed in a local private AI computing terminal that is completely physically isolated from external public networks, supporting independent offline operation within a home LAN environment, and prohibiting the transmission of raw multimodal data to the public network throughout the entire process; the system includes: The multi-source data acquisition interface layer is used to access and acquire multimodal raw data of family members at different life stages within the home LAN. The opening of the IoT hardware acquisition channel requires separate authorization from the user's liveness. The dual-track data processing engine, connected to the multi-source data acquisition interface layer, includes a parallel generative backtracking processing track and a native time-series recording track. It is used to divert the multimodal raw data to the corresponding track for differentiated processing based on the age characteristics of family members, and to achieve the same-source fusion of dual-track data through a globally unified kinship identification. The digital genealogy modeling engine, connected to the dual-track data processing engine, is used to construct multi-dimensional graphs from the fused data, generating a structured digital genealogy model that includes emotional, career, and inheritance dimensions, and storing it in a local offline graph database. The interaction module is used to receive natural language query commands from users within the local area network, route the commands to the corresponding graph dimension based on intent recognition, and drive the local large language model to generate enhanced search answers. The intergenerational inheritance module is used to configure hierarchical access permissions within the family LAN, and automatically freezes the modification permissions of the corresponding data after verifying the user's death status, switching to a read-only digital ancestor interaction mode. The privacy protection and data destruction module is used to perform hardware-level encrypted storage of all data in the terminal based on national cryptographic algorithms. The module supports dual execution scenarios: destruction delivered by service providers' field equipment and self-destruction by users' local terminals. When destruction is triggered, physical-level data erasure based on both national and international standards is performed, and the destruction report is written to a one-time read-only log area. Its hash digest can be synchronized to the consortium blockchain for evidence storage via offline physical media.

2. The system according to claim 1, characterized in that, The local private AI computing terminal is a hardware device that integrates edge heterogeneous computing power units. Its motherboard has a hard-wired hardware encryption chip that supports national cryptographic algorithms. The hardware encryption chip has a built-in true random number generator, which is used to generate, derive and manage SM4-GCM data encryption keys in the secure area inside the chip. The keys are physically prohibited from being exported to the outside of the chip.

3. The system according to claim 1, characterized in that, The specific data splitting and processing logic of the dual-track data processing engine includes: For users whose age is greater than or equal to a preset advanced age threshold, or who lack continuous native data, they are diverted to the generative retrospective processing track. The locally deployed lightweight AI model is used to complete and structure the existing historical unstructured materials, and implicit watermarks and explicit interface identifiers are forcibly embedded in the underlying layer of the AI-generated content. For natural persons whose age is less than the preset old age threshold, they are diverted to the native time-series recording track. The IoT collection channel in the local area network is dynamically started and stopped according to the preset life cycle age group. The multimodal raw time-series data is stored in lossless encryption and periodically generated with timestamp indexes. AI semantic rewriting and content extraction are prohibited. When a user in the native time-series recording track reaches the preset advanced age threshold, their objective time-series behavioral data is automatically de-identified and extracted, and then used as supplementary material for cross-track unidirectional input to the generative backtracking processing track.

4. The system according to claim 1, characterized in that, The digital genealogy modeling engine adopts an attribute graph data model, which includes three orthogonal sub-models: The emotional connection model, based on graph attention network, is used to characterize the strength of emotional relationships among family members and the individual emotional decision-making process. The sub-model, built on a temporal graph neural network, is used to represent the causal logic, success and failure attribution, and event chain of family members' career decisions. The inheritance model is used to construct a timeline map representing the geographical migration and iteration of family traditions and culture. All three orthogonal sub-models achieve cross-dimensional semantic association through the globally unified kinship identifier.

5. The system according to claim 1, characterized in that, The interaction module incorporates a lightweight intent classification model based on a deep twin network architecture, which is used to split user semantics in real time within a local area network. It adopts a GraphRAG retrieval-enhanced generation architecture to retrieve the original factual material of the three-dimensional graph, drive the local offline large model to generate answers that fit the logic of the parties involved, and forcibly attach AI-generated risk warning information at the end of the reply.

6. The system according to claim 1, characterized in that, The intergenerational transfer module is configured with four levels of differentiated intranet access permissions: read-only Q&A, image viewing, data export, and model editing; and supports full offline export of digital assets through encrypted physical media when the network is disconnected.

7. The system according to claim 1, characterized in that, The privacy protection and data destruction module executes a unified hardware-level erasure standard in the dual execution scenario: For magnetic storage media, perform three consecutive full disk overwrite operations, including full overwrite of 0x00, full overwrite of 0xFF, and true random numbers; For semiconductor storage media, an ATA underlying security erase command is issued, and the data encryption key inside the hardware encryption chip is physically erased simultaneously. The destruction report generated after the destruction is completed, which includes the device's unique identification code, operation timestamp, and liveness verification record, will be permanently stored as a local, tamper-proof credential.

8. A method for constructing a family digital genealogy based on dual-track data processing, characterized in that, Applied to the system according to any one of claims 1 to 7, the method is performed entirely in a local area network environment physically isolated from public networks, and includes the following steps: Step S1: Deploy a local private AI computing terminal in the home LAN environment, initialize the identity information of family members and generate a globally unified kinship identity mapping table, and synchronize the hash digest of the initialized data to the consortium blockchain for basic evidence storage through offline encrypted media; Step S2: Obtain the target user's age information and local native data storage, and based on the dual-dimensional judgment logic, divert the user to the generative backtracking processing track or the native time-series recording track; Step S3: In the generative backtracking process, based on the existing unstructured materials, use the local offline large model to perform structured sorting and lightweight AI completion, and embed generative watermarks; Step S4: In the native time-series recording track, the local area network dynamically activates the corresponding IoT acquisition channel, performs national cryptographic hardware-level encryption storage on the raw sensor data, and periodically closes to generate an annual memory index; when the track user reaches the preset old age threshold, their time-series data is cross-tracked to supplement the generative backtracking processing track. Step S5: Based on the globally unified kinship identity identifier, the dual-track heterogeneous data are fused together to construct a digital genealogy model in this map database that includes three-dimensional sub-maps of emotion, career, and inheritance. Step S6: Provide natural language human-computer interaction services within the local area network based on a preset hierarchical authorization mechanism, and automatically switch the account to read-only digital ancestor mode after verifying the user's death; Step S7: When the service provider completes the delivery of field materials or when the user actively initiates a clearing command, the corresponding hardware-level data destruction process is triggered, a destruction report containing a timestamp is generated and written to the local read-only log area, and the hash digest is supported to be synchronized offline to the blockchain for third-party verification.

9. The method according to claim 8, characterized in that, In the step of triggering the corresponding hardware-level data destruction process: If it is a service provider's field material delivery scenario, after the data from the field collection device is completely imported into the home LAN terminal, immediately perform a hardware-level destruction process on the field device to clear the temporarily collected sensitive data; If the user initiates the cleanup process, a hardware-level destruction process will be performed on the specified local storage medium within the home LAN terminal after authentication.

10. The method according to claim 8, characterized in that, The hardware-level data destruction process specifically includes: performing underlying security erasure and physical erasure of encryption keys on semiconductor storage media; performing multiple data overwrites on magnetic storage media; and finally writing the generated compliant destruction report into a one-time read-only log area to ensure that the destruction behavior is irreversible and traceable at the physical level.

Images