Vu Nguyen

May 2025

Try the bot here.

What’s Agak-Agak?

It’s a Singlish phrase meaning to estimate, doing guesswork. Showing a resume is like this to me, just estimating and doing guesswork. So I decided to make it a bit more fun and interactive with this bot and name it “guesswork” essentially.

How is the bot built?

Figure: Generalized architecture of Agak-Agak (source)

Core Components & Workflow

1. Document Processing and Chunking

• My resume and supplementary context documents are split into smaller text chunks using a smart splitter that respects natural language boundaries (paragraphs, sentences).
• Each chunk overlaps with the previous one to preserve context across chunk boundaries, improving semantic coherence.

2. Embedding and Vector Storage

• Each chunk is transformed into a dense vector embedding using a SentenceTransformer model (all-MiniLM-L6-v2) to capture its semantic meaning.
• These embeddings are stored in ChromaDB, an efficient in-memory vector database optimized for similarity search.

3. Query Processing and Retrieval

• When a user submits a question, the system embeds the query into the same vector space and performs a fast approximate nearest neighbor search in ChromaDB to retrieve the top candidate chunks.
• To enhance precision, the top 2× the requested results are then reranked using a cross-encoder model (ms-marco-MiniLM-L-6-v2) which jointly considers the query and each candidate chunk to produce a more accurate relevance score.

4. Answer Generation with LLM