Build a Family Health System with Doubao & Feishu: Open-Source Architecture & Automation Guide

Why Move from Single-Query to Multi-Table Closed Loop?

The author identifies three pain points with earlier approaches:

Stateless single-window Q&A : Context resets after each chat, requiring repeated history input.

Static wide-table dashboard : A 70+ column table becomes unmanageable with multiple members, batches, and reference ranges; lacks workflow (generate to-do → scheduled inspection → status closure).

Third stage: Doubao Skills + Feishu 10-table linked architecture — three engineering principles:

Zero-burden collection : Frontend only captures photos, voice, PDFs; Doubao handles multimodal recognition, field extraction, unit normalization, and auto-write.

Normalized long-table storage with dual-track binding : All structured data into one Feishu multi-dimensional table; all original PDFs/images into Feishu Drive by member. Every metric uses one row = one member, one event, one indicator long-table form, enabling cross-period trend calculation.

State-machine auto-closure : Any record marked needs follow-up auto-generates a calendar to-do, stays in a pinned filter view until done, and daily cron jobs push reminders.

System Architecture: Four-Layer Stack & Star Data Model

1. Four-Layer Technical Stack

Layer ① Collection Input : Phone photo, voice, PDF, Feishu Miaoji transcription. Engineering problem solved : Compress daily input to ≤3 seconds, lower entry friction.

Layer ② Data Foundation : Feishu multi-dimensional tables (10 linked tables) + Feishu Drive (per-member folders) + architecture handbook. Engineering problem solved : Bidirectional binding of structured details & unstructured originals, relational mapping, cross-period retrieval.

Layer ③ Intelligent Skills : Doubao 6 standardized Skills (SOP prompt library) + authoritative public knowledge sources. Engineering problem solved : Acts as smart ETL middleware between collection and DB: parsing, comparison, aggregation, write-back.

Layer ④ Application & Time : 6 high-frequency filter views + global dashboard + 5 fully automatic cron workflows. Engineering problem solved : Multi-dimensional visualization, unclosed-item tracking, periodic task triggering.

2. Feishu Multi-Dimensional Tables: 10-Table Star Topology

Centered on members (family member master table), one-way links build a star schema: members (Member Master): Name, relation, birth date, tags, personal Drive folder link. reports (Report Master): Linked member, report type, date, institution, Drive link, core conclusion, flow status ( pending parse / closed / needs follow-up). lab_indicators (Indicator Long Table): Linked report & member, indicator name, category, value, unit, ref low/high, over-limit flag, focus boolean. encounters (Communication Log): Linked member, date, department, description, conclusion, follow-up plan, Feishu Miaoji link. medications (Item Spec Table): Linked member & record, name, spec, start/end date, status ( active / stopped / standby), notes. symptoms (Daily Status Log): Linked member, status description, timestamp, trend, daily handling, flow status. vaccinations (Periodic Project Table): Linked member, project name, batch, planned date, actual date, status ( done / to schedule / overdue). checkup_schedule (To-Do Plan Table): Linked member & trigger report, plan item, target date, status ( to schedule / booked / done / overdue). family_history (Long-Term Baseline): Relative relation, background, period, long-term attention hints for family. health_goals (Annual Goals): Linked member, year, goal name, category, quantified target, current progress, achievement status.

Skills Middleware, Views & Automation Workflows

1. Six Standardized Skills Data Pipeline

To avoid LLM field omissions or format drift, high-frequency logic is encapsulated as 6 Skills (full prompts in docs/03-skills-prompts.md):

report-reader (Multimodal Report Parsing & Ingestion) : On PDF/image receipt, auto-executes full loop: create reports master record → split multi-row details into lab_indicators long table → dynamically match reference ranges & flag outliers → fetch public science references → auto-generate follow-up items in checkup_schedule.

symptom-recorder (Daily Status Structured Archival) : Converts free-text description into standard fields (subject, time, manifestation, duration) and writes to symptoms.

medication-helper (Leaflet Extraction & Deduplication) : OCR extracts public leaflet fields, cross-references members and medications existing records, assists comparison of similar entries, organizes daily reminder schedules.

visit-brief (Cross-Table Aggregation Brief & Status Closure) : Three-step data flow for offline consultations:

Pre : Auto-pulls historical over-limit indicators, recent logs, pending questions across 5 tables → generates one-page structured brief in 30 seconds.

During : Feishu Miaoji records & transcribes minutes.

Post : One sentence splits minutes back into corresponding tables, and closes completed plans in checkup_schedule.

health-review (Periodic Full-Table Review) : Monthly or yearly auto-scans all 10 tables, summarizes new records, outlier trends, unclosed items, and health_goals achievement rate.

msd-reference (Authoritative Knowledge Base Directed Retrieval) : Constrains LLM to prioritize public authoritative science sources with original URLs, reducing hallucination risk.

2. Six Filter Views + Global Dashboard + Five Cron Jobs

On top of the data foundation, Feishu views and automation enable "data finds people":

6 high-frequency filter views : Pre-configured "Unclosed Items", "To-Do Plans", "Focus Indicators", "Active Items", "Report Timeline", "Group by Member" — no raw-row hunting.

"Family Health Overview" Dashboard : Single-screen board aggregating core indicator cross-period line charts, unclosed stats cards, total report count, to-do count, monthly record stats, indicator category distribution.

5 fully automatic cron tasks :

Daily To-Do Inspection Reminder — Trigger: Daily 09:00. Logic: Scans checkup_schedule; pushes Feishu message for due/overdue items.

Monthly Data Summary Push — Trigger: 1st of month 09:00. Logic: Triggers health-review skill; auto-summarizes last month's new reports, outliers, daily logs, unclosed items.

Annual Comprehensive Check Reminder — Trigger: Jul 1 09:00. Logic: Pushes annual plan arrangement reminder; updates cross-year baseline after report ingestion.

Seasonal Cycle Prep Reminder — Trigger: Jul 15 09:00. Logic: Two-week advance trigger for seasonal focus items: stock check & schedule reminder.

Family Annual Data Review — Trigger: Dec 1 09:00. Logic: Aggregates yearly core indicator trends, calculates health_goals achievement, initializes next year's goals.

Full Open Source: One-Click Replication with Doubao

All de-identified engineering assets are in the open-source repo tjxj/family-health-system containing four core files: docs/01-design-handbook.md: Architecture handbook — four-layer design, daily/periodic/long-term time matrix, data flow specs. docs/02-table-schema.md + templates/table-schema.json: Complete 10-table field definitions, types, enum values, star one-way link mappings; human-readable Markdown + AI-consumable JSON template. docs/03-skills-prompts.md: 6 Skills standard prompt library — full SOPs and copy-paste system prompts. docs/04-user-manual.md: Daily ops manual template — one-page cheat sheet (4 core entry points + 3 high-frequency input actions + cron task list).

Three-step replication via Doubao:

Feed Schema for Auto Table Creation : Drag templates/table-schema.json and the four docs/ files into Doubao chat, send the optimized starter prompt (provided in article). Doubao auto-creates the 10-table multi-dimensional table with all fields, enums, star links, core views, dashboard; creates per-member Drive folders (with yearly report, communication, daily subfolders) and back-fills links to members; generates "6 Skills Prompt Library" and "User Manual" cloud docs.

One-Sentence Member & Cron Customization : Tell Doubao your family member names and each person's annual focus; it initializes members records, Drive subfolders, and personalized cron reminders.

Upload One Historical PDF for End-to-End Acceptance : After foundation is ready, send a past PDF report with "parse and archive" — verifies multi-table write and line-chart rendering loop.

The article includes the exact starter prompt (see source) for copy-paste use.

Summary

This "Doubao + Feishu" family data management system is the author's most satisfying personal engineering practice of the year. Facing scattered historical PDFs, dense tabular indicators, and multi-year periodic plans, traditional manual folder/Excel approaches rarely persist. Leveraging Doubao's multimodal parsing, standardized Skills orchestration, and Feishu multi-dimensional tables' relational foundation, non-programmers can spin up a structured personal/family digital archive with state closure in half an hour — no backend code required.

Note: This open-source project and demo workflow are for personal/family document archiving, data visualization, and schedule reminders only; they do not constitute professional advice.