Healthcare CRM Platform

• What it is: the core product of the Healthcare platform ecosystem — a CRM + platform for medical labs/clinics: advertisers, patients, Telegram bots, AI analysis of checkups and lab results (RAG/LLM). The legacy layer is Django; Platform v2 is migrating to Elysia/FastAPI/React/Expo (per CLAUDE.md).
• Type / status / role: web-app (multi-service medical monolith + services) · active · lead — the user is the largest author (~962 of 2208 commits, ~44%), team of ~10 (aniashev/Anna 450, Denis Ergashbaev 255, Katya Mun 248, Alisher Mukhtorov 341, Anton Kim, Bobur).
• Activity period: 2024-11-13 → 2026-01-19 — ~14 months of continuous development, the longest and largest project in the portfolio.

• Backend: Django 5.1 + django-htmx + crispy-forms + widget-tweaks, Celery 5.4 (+ beat + results) for async/periodic, gunicorn, whitenoise, PyJWT, loguru, django-stubs.
• AI/ML/RAG (key): LangChain + langchain-community, Pinecone (vector DB, gRPC), sentence-transformers (embeddings, torch/transformers), RAGAS (RAG quality eval), Langfuse (LLM observability/tracing), datasets, jinja2-fragments. → A serious production RAG/LLM pipeline (agentic checkup analysis, lab-result interpretation — agentic_checkup_analyzer.log, lab_result_worker.log).
• Services/integrations: FastAPI (healthcare-platform-core/webapp), pyTelegramBotAPI (patient bot + general bot), Firebase Admin, Google Calendar API, Sentry, PostHog, SSE (sse-starlette), PDF parsing (pymupdf, pdfplumber), thefuzz (fuzzy matching), Babel/langdetect (i18n).
• Frontend: HTMX + Alpine.js + TailwindCSS, templates by domain (advertiser/patient), strict naming conventions (list/detail/form/delete + partials).
• Infra/deploy: multi-service Docker (Dockerfile.{django,advertiser,patient,patient_telegram,telegram,webapp,workers,monitor,bitwarden-init,linting}), docker-compose (dev/prod/sa variants), Bitwarden for secrets management (init container), prod_deploy.sh, Makefile. SQLAlchemy 2.0 (in the FastAPI part), uv/pyproject.
• Data: relational DB (Django ORM), Pinecone (vectors), Firebase.

2208 commits, ~962 by the user — flagship scale. By artifacts/plans:

• CRM core: advertiser and patient portals (Django + HTMX), dashboards, forms.
• AI analysis of medical data: agentic checkup analyzer + lab-result interpretation worker (RAG: Pinecone + sentence-transformers + LangChain, eval via RAGAS, tracing via Langfuse).
• Telegram bots: patient bot + general bot (pyTelegramBotAPI), separate Docker services.
• Async infrastructure: Celery workers + beat (ACTIVITY_PLAN_CELERY_INTEGRATION, run_workers.sh).
• Payments/currency (CURRENCY_CONVERSION_IMPLEMENTATION), bulk purchases (BULK_PRODUCT_PURCHASE_PLAN), activity tracking (ACTIVITY_*).
• Secrets via Bitwarden (BITWARDEN_INIT_FIX_SUMMARY).
• Migration to Platform v2 (Elysia/FastAPI/React/Expo) — the current direction.
• Volume: 14 months, ~20 plans/docs (.md), multi-service architecture.

Confirmed by CLAUDE.md/pyproject/structure:

• Production RAG pipeline for medicine (LangChain + Pinecone + sentence-transformers + RAGAS + Langfuse): not just an "LLM call" but a full RAG with vector search, quality evaluation (RAGAS), and observability (Langfuse). Agentic checkup analyzer and lab-result interpretation. → Serious AI engineering, not a demo.
• Multi-service decomposition (10 Dockerfiles): django, advertiser, patient, two Telegram bots, webapp, workers, monitor, bitwarden-init, linting — separation by responsibility within one repo. → Mature operational architecture.
• Secrets management via Bitwarden (init container): secrets not in env files, but via the Bitwarden CLI at startup. → Security maturity.
• Async on Celery (beat + results): periodic tasks (reports, sync, AI workers) lifted out of the request cycle. → Proper handling of heavy medical workloads.
• Multilingual medicine (Babel + langdetect + transliterate): RU/UZ/localization, important for the market.
• 🌟 Advanced multi-agent Claude Code workflow (CLAUDE.md): the user built out a team of CUSTOM sub-agents — django-backend-specialist, advertiser-frontend-developer, patient-frontend-specialist, linting-specialist — with strict rules, an inter-agent communication protocol (backend reports view types / URLs / context; frontend asks them) and a fixed order "backend → frontend → linting." → A rare, strong example of a mature AI-native development process on a large codebase.

• Sessions found: 3 (directories for healthcare-crm and healthcare-platform-healthcare-crm, subdir form).
• What was done with AI: feature work per the CLAUDE.md protocol via specialized sub-agents; the huge CLAUDE.md = guide for the agent team.
• AI workflow patterns (important for the brand): multi-agent orchestration (role-specialized sub-agents + an inter-agent contract + mandatory linting agent), spec/rule-oriented development, AI workers in production (checkup analyzer). The strongest AI-workflow case among all analyzed projects.

• ~962 user commits (of 2208) over 14 months — the portfolio flagship; #1 contributor on a team of ~10.
• Multi-service platform: 10 Docker services, Django + Celery + FastAPI + 2 Telegram bots.
• Production RAG/LLM for medicine (Pinecone, RAGAS, Langfuse).
• Bitwarden secrets, Sentry/PostHog observability, Google Calendar/Firebase integrations.
• Mature multi-agent Claude Code process (4 custom sub-agents).