Township Health Agent
Policy Q&A, health records, family-doctor sign-up and check-up booking, all inside the county hospital WeChat account — nothing for residents to install.
About this page
No real resident health records, sign-up data, or live system screens are shown on this page. The system is an assistive tool for health information and policy look-up only. It does not diagnose, does not prescribe, and does not replace care delivered by a licensed medical institution — anyone feeling unwell should see a clinician. All health-insurance reimbursement and chronic-disease subsidy details follow the latest rules published by the local health-insurance authority; system answers are for reference only.
In short
The Township Health Agent is a solution design for county health groups and township clinics. It embeds an AI knowledge-base assistant, health-record intake with family-doctor sign-up, and check-up booking directly into the county hospital WeChat account, so residents look up policies, register records, and book screenings without installing anything. It is an assistive tool for health information and policy look-up — it does not diagnose, does not prescribe, and does not replace a licensed medical institution; insurance rules follow the local authority.
What the agent takes off people's plates
Core capabilities
Interactive knowledge-base Q&A (RAG)
Bulk-import policy documents and common-illness handbooks; chunks are vectorized automatically. pgvector similarity search plus keyword recall returns cited passages in a turn-by-turn dialogue. Answers come only from documents the clinic uploaded — the model does not improvise.
Health records + family-doctor sign-up
Standardized forms capture basic details, medical history, and current medication into a structured electronic record. Family-doctor sign-up happens online with a named responsible physician, status visible in real time, and the back office presents everything as tables that match the clinic's existing workflow.
Check-up booking management
Slot capacity is defined per time window; residents pick a slot and a package and book in one step. The back office aggregates by day or week and exports reports, while a cron scheduler pushes reminders before the appointment — replacing phone queues and staffing clashes.
Agent embedded in the WeChat account
DataWeaver's WeChat Channel connects to the county hospital's official-account message API. OpenID binds a resident identity so one authentication unlocks every function, and the responsive H5 pages fit phones and tablets — follow the account and start using it, nothing to install or learn.
Side by side
Township Health Agent vs. staffed counter only vs. generic chatbot
| Dimension | Township Health Agent | Staffed counter only | Generic chatbot |
|---|---|---|---|
| Consistency of policy answers | Answers are retrieved only from the policy files the clinic imported, so the same question yields the same wording for everyone; updating a file updates every answer. | Depends on each staff member's memory and experience; wording drifts between people and shifts. | Generates from general-purpose corpora and may fabricate local policy details that do not exist, with no traceable basis. |
| Retention of intake data | Forms write straight into structured PostgreSQL records; sign-up and booking status is queryable, countable, and exportable. | Paper forms and scattered spreadsheets; totals must be tallied by hand, and records are easy to lose and hard to reuse. | Chat only, nothing persisted — data disappears when the session ends, so it cannot carry a health-record workflow. |
| 7×24 availability | The WeChat entry point stays online around the clock — policy look-ups and bookings work at night and on holidays, with hard cases escalated to staff for next-day follow-up. | Only during office hours, which misses migrant workers and residents who are busy during the farming season. | Also always on, but cannot complete sign-up or booking actions that must write into a business system. |
| Traceability | Every answer cites the retrieved passage and its source document; conversations and actions are logged for later review and accountability. | Verbal answers leave no record — there is no way to reconstruct what was said or which version of a document it was based on. | No citations; the same question can get two different answers with no way to check the basis. |
How the agent and human supervision collaborate
System architecture (conceptual)
source
Touchpoints: WeChat account / H5 / clinic admin console
agent
DataWeaver gateway: WeChat Channel · AgentRouter · SessionManager · PermissionEnforcer
agent
Agent layer: knowledge Q&A / health records / booking / data query
source
Data & knowledge: pgvector store · PostgreSQL · knowledge documents · HIS integration (planned)
agent
LLM layer: Qwen 3.x / DeepSeek V4 self-hosted, Claude 5 / GPT-5 optional, multi-model fallback
human
Clinic staff: review of policy edge cases and human fallback
output
Output: cited answers · electronic health records · sign-up and booking records
Flow
- Touchpoints: WeChat account / H5 / clinic admin console→DataWeaver gateway: WeChat Channel · AgentRouter · SessionManager · PermissionEnforcerMessage ingress
- DataWeaver gateway: WeChat Channel · AgentRouter · SessionManager · PermissionEnforcer→Agent layer: knowledge Q&A / health records / booking / data queryRouting & authorization
- Agent layer: knowledge Q&A / health records / booking / data query→Data & knowledge: pgvector store · PostgreSQL · knowledge documents · HIS integration (planned)Retrieve & persist
- Agent layer: knowledge Q&A / health records / booking / data query→LLM layer: Qwen 3.x / DeepSeek V4 self-hosted, Claude 5 / GPT-5 optional, multi-model fallbackInference call
- Agent layer: knowledge Q&A / health records / booking / data query→Clinic staff: review of policy edge cases and human fallbackEscalate to staff
- Clinic staff: review of policy edge cases and human fallback→Output: cited answers · electronic health records · sign-up and booking recordsReviewed reply
- Agent layer: knowledge Q&A / health records / booking / data query→Output: cited answers · electronic health records · sign-up and booking recordsAnswer & record
Verifiable facts
Numbers and sources
Self-reported: reusing DataWeaver's existing agent framework, channel connectors, and knowledge-base engine cuts roughly 70% of duplicated development effort.
Platform components already available: KnowledgeAgent, pgvector RAG retrieval, WeChat Channel, PostgreSQL business store, Admin UI, NL2SQL querying.
HIS integration and some business-specific custom agents are planned capabilities that have not yet been delivered or validated in this solution.
Rollout runs in four steps: build the knowledge base → develop business agents → connect the WeChat account → integrate, canary, and go live.
Delivery and pricing
Pricing
Included
- · Self-hosted knowledge-base agent: RAG retrieval engine plus knowledge-base admin console
- · Health-record and family-doctor sign-up forms, plus check-up booking with slot-capacity management
- · WeChat official-account channel integration and joint testing (OpenID identity binding, message routing, responsive H5 pages)
- · Clinic admin console: sign-up and booking tables, daily/weekly statistics, report export
- · Initial knowledge-base ingestion (policy files and common-illness material, up to 200 documents) and one administrator training session
- · First year of baseline operations support and defect fixes after go-live
Provided by the client
- · HIS integration requires a separate effort assessment — it depends on how far the HIS vendor opens and supports its interfaces, and is quoted per person-day after an on-site review.
- · Infrastructure — servers, GPUs, storage, public bandwidth — and any security-grading assessment fees.
- · WeChat official-account entity qualification, verification and annual fees, plus carrier charges for SMS and push channels.
- · Ongoing curation and policy updates for the knowledge base (available separately as an annual content-maintenance service).
Add-ons
- · Technical support and version upgrades from year two, billed annually.
- · Replicated deployment to additional clinics or campuses, billed per site.
- · Custom business agents, bespoke reports, and third-party system integrations, billed per person-day.
- · Annual knowledge-base content maintenance: tracking policy updates and reviewing answer quality.
Scan for the full solution document
Scan with WeChat to reach the solution lead and request the complete documentation and deployment checklist.
Where it fits
When this approach does not apply
- The system is an assistive tool for health information and policy look-up. It does not diagnose, does not prescribe, and does not replace a consultation with a physician. Any symptom assessment, medication change, or emergency requires the resident to visit a medical institution in person — the system only points to information and to where care is available.
- Reimbursement rates, chronic-disease catalogues, annual caps, and required paperwork differ by region and by year and change as policy is revised. System answers reflect only the policy files the clinic uploaded and are for reference; the latest rules published by the local health-insurance authority and confirmed at its counter always prevail.
- Data exchange with the clinic's existing HIS is a planned capability, not part of the standard delivery. Feasibility depends on how far the HIS vendor opens its interfaces, on the on-site network, and on data-compliance approval — an on-site assessment and a separate quote are required first.
- Answer quality depends entirely on how current the knowledge base is. The clinic must assign someone to update documents whenever policy changes, otherwise the system will answer from stale files — a quarterly knowledge-base review is the minimum we recommend.
- Health records are sensitive personal data and must be deployed on servers the county controls, with consent notices, access control, and audit logging in line with personal-information protection requirements. Hosting intake data on a public-cloud generic chatbot service is not advisable.
Further reading