🤰 BabyOS — AI Pregnancy Intelligence Hub

An AI-powered pregnancy companion built with LangChain, LangGraph, ChromaDB, and Streamlit. Supports mom, dad, and partner roles with week-aware responses, RAG over medical and Germany-specific content, and multi-agent architecture.

---

Project Structure

babyos/
├── corpus/
│   ├── corpus_fetcher.py     # Step 1: Scrape NHS/WHO web content
│   └── ingest.py             # Step 2: Chunk, embed, store in ChromaDB
├── data/
│   ├── raw/                  # Source documents (markdown, json, scraped txt)
│   │   ├── fetal_development.json
│   │   ├── medical_reference.md
│   │   ├── germany_pregnancy_guide.md
│   │   ├── dad_partner_guide.md
│   │   ├── pregnancy_faqs.md
│   │   └── web_scraped/      # Created by corpus_fetcher.py
│   └── chroma_db/            # Created by ingest.py
├── rag/
│   └── rag_system.py         # BabyOSRetriever + RAG chain factory
├── agents/                   # (Day 3) LangGraph multi-agent system
├── utils/                    # (Day 4) User profile, session state
├── app.py                    # (Day 4) Streamlit UI entry point
├── .env.example              # Copy to .env and fill in keys
└── requirements.txt

---

ChromaDB Collections

============================================================ INGESTION SUMMARY — babyos_universal

Total chunks: 11263

By source type: pdf 8719 youtube 2032 web 412 markdown 88 json 12

By topic (top 15): medical_board 2844 ██████████████████████████████ labor 1949 ██████████████████████████████ diet_advice 1728 ██████████████████████████████ informed_choices 889 ██████████████████████████████ pregnancy_symptoms 771 ██████████████████████████████ pregnancy_workout 770 ██████████████████████████████ fetal_movement 646 ██████████████████████████████ mental_health 606 ██████████████████████████████ clothing 383 ██████████████████████████████ first_weeks 198 ██████████████████████████████ for_partner 169 ██████████████████████████████ baby_development 117 ███████████████████████ breastfeeding 102 ████████████████████ baby_care_guide 91 ██████████████████

By period (pregnancy months): pregnancy_m1 29 pregnancy_m2 104 pregnancy_m3 160 pregnancy_m4 103 pregnancy_m5 52 pregnancy_m6 114 pregnancy_m7 96 pregnancy_m8 29 pregnancy_m9 1568

By period (postpartum, key months): postpartum_m0 590 postpartum_m1 6 postpartum_m3 53 postpartum_m6 60 postpartum_m12 70 postpartum_m24 9 all 8173

---

Quickstart

1. Install dependencies

python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Set up environment

cp .env.example .env
# Edit .env and add your OPENAI_API_KEY (and optionally LANGCHAIN_API_KEY)

3. Fetch web corpus (optional — adds NHS/WHO content)

python corpus/corpus_fetcher.py

This takes ~5 minutes and is polite to the servers (1.5s delay between requests). You can skip it and ingest only the local files first.

4. Run ingestion

# First time
python corpus/ingest.py

# Reset and rebuild from scratch
python corpus/ingest.py --reset

Expected output:

babyos_development          12 chunks
babyos_medical              28 chunks
babyos_germany              22 chunks
babyos_dad                  31 chunks
babyos_faqs                 19 chunks
babyos_web                 180 chunks   (if fetched)
TOTAL                       292 chunks

5. Smoke test the RAG system

python rag/rag_system.py

6. Run the app (once built)

streamlit run app.py

---

Architecture

User Input
    │
    ▼
Streamlit UI (app.py)
    │
    ▼
LangGraph Supervisor
    │
    ├── Medical Q&A Agent      ──► babyos_medical + babyos_web
    ├── Baby Tracker Agent     ──► babyos_development
    ├── Emotional Support Agent──► user context + memory
    ├── Dad/Partner Agent      ──► babyos_dad + babyos_development
    └── Germany Info Agent     ──► babyos_germany
         │
         ▼
    BabyOSRetriever (rag_system.py)
         │
         ▼
    ChromaDB (6 collections)
         │
         ▼
    GPT-4o / GPT-4o-mini
         │
         ▼
    LangSmith (tracing + evaluation)

---

Deployment

Streamlit Cloud (recommended for demo)

1. Push repo to GitHub
2. Go to share.streamlit.io → New app
3. Add secrets in Streamlit Cloud dashboard (copy from .env)
4. Deploy

HuggingFace Spaces

• Use Streamlit SDK
• Note: ChromaDB persistence requires paid storage; use in-memory for demo

Docker

docker build -t babyos .
docker run -p 8501:8501 --env-file .env babyos

---

LangSmith Evaluation

Set LANGCHAIN_TRACING_V2=true in your .env.
All RAG retrievals and agent responses are automatically traced.

To run evaluations:

python eval/run_evals.py    # (created on Day 5)

---

Important Notes

• This app is not a substitute for professional medical advice. Every response from the Medical Q&A agent includes a disclaimer.
• Danger signs trigger an immediate escalation response recommending the user contact their doctor or call emergency services.
• Scan image analysis uses GPT-4o Vision for descriptive summaries only — it does not provide diagnostic interpretations.