06 — Synthetic Research · UX Intelligence
Explore research hypotheses before fieldwork.
A unified research workspace that turns a study brief into a diverse synthetic cohort, simulated qualitative responses and an actionable synthesis. When a screen is provided, visual-attention analysis can enrich the readout.
01 — THE CHALLENGE
Product teams often lose context between research framing, participant recruitment, screen evaluation and synthesis. This project asks how these steps can be connected in one reproducible workflow while keeping synthetic evidence clearly distinct from real user research.
Project anatomy
A rapid technical read connected to visible product features and grounded in the architecture currently implemented.
Core pipeline
Select a step to understand how the data is transformed.The researcher defines product context, objective, target users, questions and cohort size in the Next.js interface.
Implemented features
Visible capabilities are connected to the technical components that make them possible.Product context, research objective, target description and custom questions.
Selection of 5, 10, 20 or 50 profiles with optional variation in the candidate pool.
Semantic search across encoded personas stored in ClickHouse.
Interview simulation informed by each persona’s behavioral profile and stance.
Structured screenshot analysis: hierarchy, visible actions, confidence cues and potential friction.
Langfuse instrumentation across the pipeline and model calls for run-level observability.
Stack by responsibility
Each technology is placed within its role in the system.TypeScript, App Router and cohort exploration components.
Brief validation, generation endpoint and typed JSON contract.
Target and persona embeddings, batch encoding and cosine proximity.
Float32 vectors, persona metadata and top-N retrieval.
Persona-level qualitative answers and optional multimodal analysis.
Pipeline traces, study metadata and model-call monitoring.
Define a target group and research objective, then follow profile selection, response simulation and the emergence of initial findings to validate in the field.
Educational demonstration: data is local and some computations are simulated or accelerated.04 — OUTCOME AND IMPACT
“A research operating system that accelerates early exploration, preserves behavioral diversity and helps teams decide what to validate with real users next.”