An New Little Project

  • Wed, May 15, 2024
  • One-minute read

Here’s my experience transitioning into AI development:

At MDAP, I was tasked with building a legal document interpret system using LLMs. This required: - Understanding generative AI basics - Learning LangChain, HuggingFace, Streamlit - Modifying third-party code

The last point was my biggest challenge. While I was comfortable modifying our Django codebase, altering LangChain’s code felt more intimidating due to its widespread use.

Initially tried:

  1. Creating workarounds
  2. Bypassing abstraction layers

Both approaches failed - workarounds created bugs, and bypassing made prototyping difficult.

The solution? Diving directly into LangChain’s source code. After studying the documentation and codebase, implementing our features was surprisingly straightforward.

Key lesson: The fear of modifying established code was more intimidating than the actual technical challenge.

– UTQ –

Posts in this series