AI Engineer · Retrieval systems · Agentic workflows
I build AI applications where retrieval, reasoning, and software engineering meet.
Models are components. Reliable systems are engineered.
I’m interested in the parts of AI engineering that make an LLM application dependable beyond a demo:
- Retrieval: hybrid search, reranking, source attribution, and context quality.
- Agentic systems: tool-using workflows with explicit boundaries and human approval.
- Evaluation: test cases and retrieval metrics that make changes measurable.
- AI backends: APIs, streaming, data stores, and the operational glue around models.
A document question-answering system that combines dense retrieval with BM25, fuses rankings, reranks candidates, and returns answers grounded in cited source passages. Includes an evaluation path with RAGAS.
ChromaDB · FastEmbed · BM25 · RRF · FlashRank · FastAPI · RAGAS
An enterprise workflow automation platform that connects LLM-driven agents to structured operations, with role-aware access and human approval steps.
FastAPI · LangGraph · DuckDB · Typst · Server-Sent Events
| Area | Tools |
|---|---|
| LLM applications | LangChain, LangGraph, tool calling, prompt guardrails |
| Retrieval | ChromaDB, FastEmbed, BM25, RRF, FlashRank, Qdrant |
| Evaluation | RAGAS, pytest |
| Backend | Python, FastAPI, Pydantic, Uvicorn |
| Storage and delivery | DuckDB, Docker, GitHub Actions |
| Model access | OpenAI, Gemini |
- A useful answer should show where its evidence came from.
- An agent should have a clear tool boundary and a safe handoff when judgment belongs to a person.
- Evaluation should describe the behavior we want to preserve—not just produce a score.
- Email: daffaarigo02@gmail.com
- GitHub: @daffaarigoh
Built around a simple idea: make the reasoning path easier to inspect.