Agentic AI Career Assistant
Goal
Build an interactive, agentic AI career assistant that answers questions about professional experience, projects, and technical skills using personal knowledge sources, while capturing visitor inquiries and continuously improving through retrieval evaluation.
Process
I wanted to make exploring my professional background more interactive than browsing a traditional portfolio. That idea evolved into an Agentic AI Career Assistant that lets visitors ask conversational questions about my experience, projects, and technical skills. The assistant uses LLM APIs and tool calling to support context-aware conversations and integrate external tools. It can log unanswered questions for follow-up, send notifications, and record contact details when visitors choose to share them. A Gradio interface brings these capabilities together into an interactive experience. Beyond the chatbot itself, I focused on making its knowledge retrieval measurable and reproducible. I implemented a semantic retrieval baseline using OpenAI embeddings and FAISS, then extended the evaluation pipeline with hybrid retrieval using BM25 keyword matching and Reciprocal Rank Fusion (RRF). I built a relevance-labeled evaluation benchmark and measured Hit Rate, Precision, Recall, and Mean Reciprocal Rank at different retrieval cutoffs. The experiments showed that hybrid retrieval improved ranking quality and recall at k=8, while also revealing a key limitation: broad questions can still surface generic profile summaries instead of specific project evidence. Rather than treating retrieval as a solved problem, I'm using these findings to guide the next iteration: query-aware retrieval, improved evidence selection, and more reliable grounded answers. The project reflects my approach to AI engineering: building useful systems, evaluating their behavior, and iterating based on measurable results.