Skip to content
Youlong MaSoftware engineerLet's talk
All work

Project 02 / Search + retrieval

Agentic Doc Retrieval

Finding the right signal in a public archive.

Contribution

Project development

Explore the original
Project documentation Source code
Study 02 / Abstract interpretationAn offset archive resolves into a highlighted document and its connected evidence.

01 / Context

The starting point.

An archive is useful when people can find the documents that matter to their question. This project focuses on FDA product recalls, making that collection searchable in natural language.

02 / Approach

How the pieces connect.

The system combines retrieval-augmented generation with LangChain agents. Vector embeddings provide the retrieval foundation, with LangGraph and OpenAI in the Python toolchain.

  1. 01Question
  2. 02Vector retrieval
  3. 03Recall documents
  4. 04Agent response
A conceptual retrieval flow. The source repository is the reference for implementation details.

03 / In focus

From question to relevant documents

The engineering focus is the connection between an unstructured question and a relevant set of recall documents. Retrieval gives the agent a document-grounded starting point for its response.

  • Python
  • LangChain
  • LangGraph
  • OpenAI
  • RAG
  • Vector Embeddings

04 / Outcome

What came together.

The existing project description reports sub-second response times across recall data dating back to 2009. The source repository provides the implementation and project documentation.

Read the project documentation
Next connection / 03AI Stock Agent