A trustworthy Swahili conversational AI for maternal and newborn health.
# MamaCare AI
MamaCare AI is a Swahili conversational AI project focused on providing
natural, accessible, and trustworthy maternal and newborn health information
via a Telegram bot, grounded in vetted, reputable health sources.
> MamaCare AI is an educational and technical project. It does not replace
> qualified medical professionals or emergency medical care.
## Repository layout
```
data/ Source register + the pipeline stages that turn vetted
sources into machine-usable knowledge (see data/*/README.md)
backend/
modules/
pipeline/ collect (web/PDF) → detect language → translate to Swahili
→ human review → publish. Feeds knowledge/.
ingestion/ absorbed into pipeline/ — see its README
knowledge/ chunk & embed approved Swahili content
storage/ vector store + runtime relational data access
rag/ retrieval + grounded generation
safety/ disclaimers, emergency detection, scope refusal, logging
api/ FastAPI routers (/chat, /health)
media/ OUT OF SCOPE for the 8-week MVP — see its README
bot/ the Telegram bot — the actual user-facing product
core/ config loading, LLM client wrapper
config/ environment settings (.env, never committed)
shared/ cross-cutting utilities
tests/ automated tests (required for every module change)
alembic/ DB migrations, only if/when needed
eval/ evaluation harness: test questions + dated reports
docs/ architecture, team, collaboration, and the 8-week program
logbooks/ each trainee's weekly work log
```
Start here, in order:
- docs/TEAM.md (who's building this and
how tracks get assigned)
- docs/ARCHITECTURE.md (how
the pieces fit together and why)
- CONTRIBUTING.md (the
day-to-day workflow: branches, PRs, Definition of Done)
- docs/COLLABORATION.md (how four people build this
independently without colliding).
- docs/INTERNSHIP_PROGRAM.md
has the full 8-week plan
- docs/DECISIONS.md is the
running log of …