“AI is only powerful if it reaches the people who need it most.”
It is with great pleasure that I announce to all that I’ve been selected as a Cohere Labs Catalyst Grant recipient!
This grant supports bold ideas that use large language models to tackle real-world problems. With Cohere’s support and $2,000 in API credits, I’m launching a project I deeply care about: BioAid QA, an open-source, multilingual biomedical question-answering assistant for underserved communities.
What is BioAid QA?
BioAid QA is a retrieval-augmented generation (RAG) assistant designed to help healthcare workers and patients access trustworthy medical information in English, Hausa, Swahili, and French.
It integrates Cohere’s Command, Embed, and Rerank models with a curated corpus of WHO and NIH resources. The system will be deployed via a lightweight Gradio interface and hosted on Hugging Face Spaces, making it accessible even in low-bandwidth settings.
Technical Overview
BioAid QA will be built as follows:
1. Document Ingestion & Indexing
- Curate biomedical texts (e.g., WHO guidelines, NIH fact sheets).
- Split into passages and embed using Cohere Embed.
- Store embeddings in a Chroma index.
2. Multilingual Retrieval & Reranking
- User questions (in multiple languages) are embedded and matched via nearest-neighbor search.
- Passages are reranked using Cohere Rerank for relevance.
3. Answer Generation
- Top passages + query → Cohere Command → concise, fact-based answers.
4. Deployment
- Gradio frontend with language toggles.
- Client-side caching and optimized UX for slow networks.
Research & Deliverables
This project doubles as software and a research contribution. The planned deliverables include:
- A peer-reviewed paper describing architecture, evaluation, and findings.
- An MIT licensed open-source GitHub repo
- A deployed app on Hugging Face Spaces
Timeline (July 2025 – January 2026)
| Month | Milestone |
|---|---|
| August | Corpus curation, preprocessing, embedding & indexing |
| September | Retrieval + rerank integration, validation with QA pairs |
| October | Command-based generation + accuracy evaluation |
| November | Refinement, paper drafts |
| December | Gradio app deployment + UX testing |
| January | Final paper submission + open-source release |
Thank You, Cohere Labs
Big thanks to Cohere Labs and the Catalyst team for supporting this project. I’m also grateful to the community at Arewa Data Science Academy for their continued encouragement.
Follow Along
I’ll be documenting the full journey — wins, failures, and findings — right here on my Quarto blog.
Stay tuned, and let’s build tools that make health knowledge accessible for everyone.
— Lukman Aliyu
lukman.j.aliyu@gmail.com