Build the evidence-first retrieval and reasoning core powering RAIVA across regulated industries.
About the role
You will own large parts of our multi-agent retrieval, evidence verification, and consortium scoring stack. We care about provable answers — not vibes — so your work directly determines whether a legal, financial, or real‑estate professional can trust an answer enough to act on it.
What you'll do
- Design and ship multi-agent retrieval pipelines (SQL, BM25, vector, optional graph)
- Own embedding strategy across primary (Cohere v4) and domain‑specific models
- Improve evidence extraction, verification, and consortium scoring quality
- Define evals and guardrails so quality regressions are caught before users see them
- Collaborate with backend/MLOps to take experiments to stable, observable production
What we look for
- 5+ years building production AI/ML or applied LLM systems
- Deep practical experience with RAG, hybrid search, embeddings, and rerankers
- Strong Python; comfortable with FastAPI, async workloads, and vector DBs (Qdrant)
- Track record of shipping evidence-backed or high‑stakes AI features
Nice to have
- Experience with Cohere, OpenAI, Gemini, or domain‑tuned embedding models
- Prior work in legal, financial, tax, or real‑estate tech
- Open‑source contributions in retrieval/LLM tooling
Stack & tools
Python
FastAPI
Supabase
Qdrant
LangChain
Cohere
OpenTelemetry
Docker
