Personal Portfolio Project
Traditional portfolio websites are static and passive. Recruiters have to:
Navigation Challenges:
Result: Recruiters may miss key accomplishments or skills that are buried in lengthy descriptions, or give up before finding what they need.
Built a production RAG (Retrieval-Augmented Generation) system that makes my portfolio interactive and conversational:
RAG Architecture:
Technical Implementation:
Evolution Journey:
Created an always-on interactive portfolio experience:
User Experience:
Technical Achievement:
Business Value:
Knowledge base: 200+ manually curated entries covering 10+ years of professional experience. Semantic embeddings: 1536-dimension vectors from OpenAI text-embedding-3-small. Response quality validated through manual testing of 50+ common recruiter questions.
RAG architecture chosen over fine-tuning for instant knowledge updates and better factual accuracy. Claude Haiku 4.5 selected for 10x faster responses and 5x lower cost vs GPT-4 while maintaining accuracy. Streaming responses for better perceived performance. Server-side caching reduces API costs by 80%.
Response time measured via server logs (< 2 seconds p95). Accuracy validated through manual review of responses to common questions. User engagement tracked via analytics (session duration, questions asked). Cost optimized through caching and efficient prompt design.
RAG latency acceptable (< 2s vs instant static content). Knowledge base requires manual curation (vs automated scraping). Claude context limits require careful prompt engineering. Streaming adds complexity but improves UX. No user feedback mechanism yet (planned for next iteration).
Solo Project - Handled all aspects:
Key Technical Decisions:
RAG vs Fine-Tuning:
Claude Haiku vs GPT-4:
Streaming vs Batch:
Technical & Business Constraints:
vs Static Portfolio Sites:
vs LinkedIn/Resume:
Key Differentiator: This portfolio demonstrates RAG implementation skills while solving a real problem (making my experience discoverable).
Key Learnings:
Knowledge Engineering is Everything: The quality of RAG responses depends entirely on knowledge base structure. I learned to:
What Worked Well:
What I'd Do Differently:
Solo Developer & Product Designer
Domain: AI/ML & Portfolio Innovation