Context
Recruiters often ask similar questions, but answers vary depending on framing, memory, or time. I wanted a single, consistent way for people to explore my experience—without exaggeration, unverified metrics, and unsupported claims.
AI Systems
RAG
Guardrails
Cost-Aware
Responsible AI
GenAI Portfolio Assistant
An evidence-based assistant embedded in this site, designed to answer recruiter questions about my experience and projects using curated knowledge, retrieval-style patterns, and guardrails that avoid unsupported claims.
Constraints
- Truthfulness: No invented metrics, titles, or responsibilities.
- Bias transparency: Positive framing is allowed, unsupported claims are not.
- UX: Fast, skimmable, and low-friction.
- Maintainability: Content must evolve without rewriting the UI.
My Role
I defined the assistant’s purpose, boundaries, and content strategy, then implemented the UI and API integration. My focus was designing behavior that prioritizes accuracy, restraint, and recruiter trust.
Key Decisions
- Evidence over impressiveness: Optimize for defensible answers, not maximal confidence.
- Explicit limits: The assistant should hedge or refuse when information is missing.
- Simple UI: A clean chat experience beats feature-heavy interfaces for this use case.
- Separation of concerns: Public narrative lives in the site; curated knowledge and retrieval documents support the assistant.
Outcome
- A live assistant embedded directly in the portfolio.
- Consistent, grounded answers aligned to real experience.
- Curated knowledge and retrieval documents that can be rebuilt into the RAG index.
Tools Used
Next.js (App Router) · TypeScript · OpenAI API · RAG patterns · shadcn/ui