Tibbiy CRM platformasi

• Bu nima: Healthcare platform ekotizimining asosiy mahsuloti — tibbiy laboratoriyalar/klinikalar uchun CRM + platforma: reklama beruvchilar, bemorlar, Telegram botlari, checkup va tahlil natijalarini AI orqali tahlil qilish (RAG/LLM). Legacy qatlami Django'da; Platform v2 esa Elysia/FastAPI/React/Expo'ga ko'chirilmoqda (CLAUDE.md bo'yicha).
• Turi / holati / roli: web-app (ko'p xizmatli tibbiy monolit + xizmatlar) · active · lead — foydalanuvchi eng yirik muallif (2208 kommitdan ~962 tasi, ~44%), ~10 kishilik jamoa (aniashev/Anna 450, Denis Ergashbaev 255, Katya Mun 248, Alisher Mukhtorov 341, Anton Kim, Bobur).
• Faoliyat davri: 2024-11-13 → 2026-01-19 — ~14 oy uzluksiz ishlab chiqish, portfoliodagi eng uzoq va eng yirik loyiha.

• Backend: Django 5.1 + django-htmx + crispy-forms + widget-tweaks, async/davriy vazifalar uchun Celery 5.4 (+ beat + results), gunicorn, whitenoise, PyJWT, loguru, django-stubs.
• AI/ML/RAG (asosiy): LangChain + langchain-community, Pinecone (vektor DB, gRPC), sentence-transformers (embeddinglar, torch/transformers), RAGAS (RAG sifatini baholash), Langfuse (LLM observability/tracing), datasets, jinja2-fragments. → Jiddiy production RAG/LLM pipeline (agentli checkup tahlili, laboratoriya natijalarini talqin qilish — agentic_checkup_analyzer.log, lab_result_worker.log).
• Xizmatlar/integratsiyalar: FastAPI (healthcare-platform-core/webapp), pyTelegramBotAPI (bemor boti + umumiy 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, domen bo'yicha shablonlar (advertiser/patient), qat'iy nomlash konventsiyalari (list/detail/form/delete + partiallar).
• Infra/deploy: ko'p xizmatli Docker (Dockerfile.{django,advertiser,patient,patient_telegram,telegram,webapp,workers,monitor,bitwarden-init,linting}), docker-compose (dev/prod/sa variantlari), maxfiy kalitlarni boshqarish uchun Bitwarden (init konteyner), prod_deploy.sh, Makefile. SQLAlchemy 2.0 (FastAPI qismida), uv/pyproject.
• Ma'lumotlar: relatsion DB (Django ORM), Pinecone (vektorlar), Firebase.

2208 kommit, ~962 tasi foydalanuvchidan — flagman miqyos. Artefakt/rejalar bo'yicha:

• CRM yadrosi: reklama beruvchi va bemor portallari (Django + HTMX), dashboardlar, formalar.
• Tibbiy ma'lumotlarni AI tahlili: agentli checkup analyzer + laboratoriya natijalarini talqin qiluvchi worker (RAG: Pinecone + sentence-transformers + LangChain, RAGAS orqali baholash, Langfuse orqali tracing).
• Telegram botlari: bemor boti + umumiy bot (pyTelegramBotAPI), alohida Docker xizmatlari.
• Async infratuzilma: Celery workerlari + beat (ACTIVITY_PLAN_CELERY_INTEGRATION, run_workers.sh).
• To'lovlar/valyuta (CURRENCY_CONVERSION_IMPLEMENTATION), ommaviy xaridlar (BULK_PRODUCT_PURCHASE_PLAN), faoliyat kuzatuvi (ACTIVITY_*).
• Bitwarden orqali maxfiy kalitlar (BITWARDEN_INIT_FIX_SUMMARY).
• Platform v2'ga (Elysia/FastAPI/React/Expo) migratsiya — joriy yo'nalish.
• Hajm: 14 oy, ~20 reja/hujjat (.md), ko'p xizmatli arxitektura.

CLAUDE.md/pyproject/tuzilma bilan tasdiqlangan:

• Tibbiyot uchun production RAG pipeline (LangChain + Pinecone + sentence-transformers + RAGAS + Langfuse): shunchaki "LLM chaqiruvi" emas, balki vektor qidiruvi, sifatni baholash (RAGAS) va observability (Langfuse) bilan to'liq RAG. Agentli checkup analyzer va laboratoriya natijalarini talqin qilish. → Jiddiy AI-injiniring, demo emas.
• Ko'p xizmatli dekompozitsiya (10 ta Dockerfile): django, advertiser, patient, ikkita Telegram boti, webapp, workers, monitor, bitwarden-init, linting — bitta repo ichida javobgarlik bo'yicha ajratish. → Yetuk operatsion arxitektura.
• Bitwarden orqali maxfiy kalitlarni boshqarish (init konteyner): kalitlar env fayllarda emas, balki ishga tushirishda Bitwarden CLI orqali olinadi. → Xavfsizlik yetukligi.
• Celery'da async (beat + results): davriy vazifalar (hisobotlar, sinxronizatsiya, AI workerlari) so'rov siklidan chiqarib olingan. → Og'ir tibbiy yuklamalar bilan to'g'ri ishlash.
• Ko'p tilli tibbiyot (Babel + langdetect + transliterate): RU/UZ/lokalizatsiya, bozor uchun muhim.
• 🌟 Ilg'or ko'p agentli Claude Code vorkflowi (CLAUDE.md): foydalanuvchi maxsus sub-agentlar jamoasini qurdi — django-backend-specialist, advertiser-frontend-developer, patient-frontend-specialist, linting-specialist — qat'iy qoidalar, agentlararo aloqa protokoli (backend view turlari / URL'lar / kontekstni xabar qiladi; frontend ulardan so'raydi) va qat'iy "backend → frontend → linting" tartibi bilan. → Katta kod bazasida yetuk AI-native ishlab chiqish jarayonining kam uchraydigan, kuchli namunasi.

• Topilgan sessiyalar: 3 (healthcare-crm va healthcare-platform-healthcare-crm uchun kataloglar, subdir shakli).
• AI bilan nima qilingan: CLAUDE.md protokoli bo'yicha ixtisoslashgan sub-agentlar orqali feature ishlari; ulkan CLAUDE.md = agentlar jamoasi uchun qo'llanma.
• AI vorkflow patternlari (brend uchun muhim): ko'p agentli orkestratsiya (rol bo'yicha ixtisoslashgan sub-agentlar + agentlararo shartnoma + majburiy linting agenti), spetsifikatsiya/qoidaga yo'naltirilgan ishlab chiqish, productionda AI workerlari (checkup analyzer). Tahlil qilingan barcha loyihalar orasida eng kuchli AI-vorkflow keysi.

• ~962 foydalanuvchi kommiti (2208 tadan) 14 oy davomida — portfolio flagmani; ~10 kishilik jamoada #1 hissa qo'shuvchi.
• Ko'p xizmatli platforma: 10 ta Docker xizmati, Django + Celery + FastAPI + 2 Telegram boti.
• Tibbiyot uchun production RAG/LLM (Pinecone, RAGAS, Langfuse).
• Bitwarden maxfiy kalitlari, Sentry/PostHog observability, Google Calendar/Firebase integratsiyalari.
• Yetuk ko'p agentli Claude Code jarayoni (4 ta maxsus sub-agent).