Technical reports for MAM-AI — an on-device medical RAG system for nurses and midwives in Zanzibar.
# MAM-AI Tech Report
LaTeX source for two arXiv tech reports documenting the MAM-AI project — an on-device medical retrieval-augmented generation system for nurses and midwives in Zanzibar.
## Papers
- **`system-paper/`** (arXiv:2606.29580) — *MAM-AI: An On-Device Medical Retrieval-Augmented Generation System for Nurses and Midwives in Zanzibar.* The deployed system, the bottom-up evaluation methodology, and headline results across retrieval, faithfulness, end-to-end accuracy, and on-device latency.
- **`benchmarks-paper/`** (arXiv:2606.29467) — *mamabench and mamaretrieval: Benchmarks for Evaluating Medical Retrieval-Augmented Generation in Maternal, Neonatal, and Reproductive Health.* The two benchmark artifacts used to evaluate the system above, with construction methodology and validation results.
Both papers are written in ACM journal style (`acmart` class, `acmsmall` variant, `nonacm` mode) and posted on arXiv. No specific venue is pursued.
## How to cite
The author's name is in Chinese order — family name **Ren**, given name **Yi**. The double braces around `{{Ren Yi}}` keep the full name as one unit so author–year styles render "[Ren Yi 2026]" and do not reorder it to "Yi Ren" or abbreviate it to the bare family name.
```bibtex
@misc{ren2026mamai,
author = {{Ren Yi}},
title = {{MAM-AI}: An On-Device Medical Retrieval-Augmented Generation System for Nurses and Midwives in Zanzibar},
year = {2026},
eprint = {2606.29580},
archivePrefix = {arXiv},
primaryClass = {cs.CL}
}
@misc{ren2026mamabench,
author = {{Ren Yi}},
title = {{mamabench} and {mamaretrieval}: Benchmarks for Evaluating Medical Retrieval-Augmented Generation in Maternal, Neonatal, and Reproductive Health},
year = {2026},
eprint = {2606.29467},
archivePrefix = {arXiv},
primaryClass = {cs.CL}
}
```
## Design
See `paper_design.md` for:
- the two-paper split rationale (why not one monolithic report, not three separate reports);
- …