Interactive RAG-powered Q&A over my published research, built with Cohere's retrieval stack. Ask questions about individual papers or synthesize themes across my body of work — all grounded with inline citations.
| Stage | Model | Role |
|---|---|---|
| Embed | embed-v4.0 |
Dense vector representations of paper chunks |
| Rerank | rerank-v4.0-pro |
Precision re-ordering of retrieval candidates |
| Chat | command-a-03-2025 |
Grounded answers with inline citations via RAG |
- Research papers are split into overlapping text chunks and embedded once (cached to disk).
- At query time, the question is embedded and matched against chunks via cosine similarity.
- Top candidates are reranked for precision, then passed as RAG documents to Cohere Chat.
- The model answers using only the retrieved context, with citations pointing back to specific papers.
- 📖 Paper Q&A — Ask any question. Relevant passages are retrieved, reranked, and used as context for a grounded, cited answer.
- 🔗 Cross-Paper Synthesis — Select 2+ papers and discover shared themes, contrasting methods, and overarching narratives. Chunks are retrieved from each paper for balanced cross-document reasoning.
| Paper | Year | Link |
|---|---|---|
| Spectral Mapping of Singing Voices: U-Net-Assisted Vocal Segmentation | 2024 | arXiv |
| On Using Quasirandom Sequences in ML for Model Weight Initialization | 2024 | arXiv |
| Lock Prediction for Zero-Downtime Database Encryption | 2025 | arXiv |
| On Sequence-to-Sequence Models for Automated Log Parsing | 2026 | arXiv |
- Python 3.13+
- uv package manager
- A Cohere API key (free trial works)
git clone https://github.com/mbrotos/cohere-explore.git
cd cohere-explore
# Install dependencies
uv sync
# Add your API key
cp .env.example .env
# Edit .env and set CO_API_KEY=your_key_here
# Run the app
uv run streamlit run app.pyThe app opens at http://localhost:8501. On first launch, document chunks are embedded via the Cohere API (~3-4 min on a trial key due to rate limits). Embeddings are cached to .cache/ — subsequent launches are instant.
A standalone script to verify your Cohere setup:
uv run python helloworld.pyDeployed via Streamlit Community Cloud with auto-deploy on push to main. The CO_API_KEY is stored in Streamlit secrets.