GenAI Insighter
Architecture of an agentic system for qualitative and quantitive research analysis (IDI / FGI / CAWI), built on deterministic tools rather than free-form generation.
- role
- Head Architect
- period
- 2026 – present
Problem
Qualitative research produces structured and unstructured data that analysts work through slowly and inconsistently. Free-form LLM analysis is neither reproducible nor cost-bounded.
What was built
- High-level design for an agentic system analysing interview and survey data.
- Deterministic tools exposed to the agent instead of free-form generation — reproducible results and control over the reasoning path with deterministic results.
- Embeddings and RAG-like retrieval — clustering and surfacing the relevant fragments of the data.
- Internal-state tracking, with guards against runaway cost and loops.
- Full cost tracking of model calls down to a single analysis and sub-agent invocation.
Outcome
- Delivered the HLD, a PoC, and the first tool-using agent, with a detailed implementation plan.
- Approved by the client’s architecture council for development and rollout across the organisation.
- Started implementation of first version.