A portfolio website featuring an AI chatbot that lets visitors ask questions about my experience, skills, and projects using RAG-powered semantic search over my professional knowledge base.
Status
Active
Role
Designer and full-stack engineer for the portfolio, retrieval pipeline, and deployment
Duration
2025
Ownership
End-to-end engineering
Problem and constraints
Recruiters and hiring managers needed a faster way to find evidence about specific skills and projects than navigating a static portfolio, while generated answers had to remain grounded in approved professional information.
Public-facing accuracy
Serverless frontend
Cold-starting AI backend
Responsive and accessible interaction
System flow
01Receive a question in the Next.js interface
02Retrieve relevant passages from the curated knowledge base
03Generate a grounded response with the selected model
04Stream the answer and preserve conversation context
Engineering decisions
Ground answers in curated evidence
Retrieval limits the assistant to approved resume and project context rather than relying on unsupported model knowledge.
Separate static content from AI availability
The portfolio remains useful when the assistant backend is cold, unavailable, or rate limited.
Evaluation
Grounded answer review
Responses are evaluated against representative recruiter questions for source relevance, factual consistency with the knowledge base, and appropriate uncertainty when evidence is absent.
Reliability, safety, and observability
Request cancellation and timeouts prevent stale responses from overwriting newer conversations.
Availability states and fallback content keep the site usable while the backend starts.
Outcomes
Deployed a responsive portfolio with a conversational evidence interface
Kept core professional content accessible independently of model availability
Failures and lessons
AI availability must not block access to portfolio evidence.
Retrieval quality and explicit evidence boundaries matter more than fluent unsupported answers.