Automated Coder Application (4.5)
About this lesson
This video extends the ChatConversation framework into a full two-agent workflow called the Coder Assistant. Instead of a single LLM, the notebook sets up both a code generator and a code reviewer, each constrained by strict rules. The generator is tasked with producing Python code for a given problem—in this case, solving the Traveling Salesman Problem with dynamic programming—while the reviewer checks the output against a detailed rubric covering correctness, complexity, edge cases, testability, readability, and performance. The reviewer signals acceptance only by returning the exact token , ensuring that only production-ready code passes . The orchestration loop manages back-and-forth between the two agents. It extracts fenced code blocks from the generator, repairs syntax errors if needed, and enforces reviewer discipline to prevent it from writing code itself. Iterations continue until the reviewer accepts the solution, the maximum round limit is reached, or stagnation is detected. The notebook also supports saving and reloading agent state, customizing verbosity, iteration limits, and memory strategies. This demonstration shows how collaborative AI agents can iteratively generate, critique, and refine solutions—scaling LangChain conversations into a structured automated coding assistant Code for This Video: https://github.com/jeffheaton/app_generative_ai/blob/main/t81_559_class_04_5_coder.ipynb ~~~~~~~~~~~~~~~ COURSE MATERIAL ~~~~~~~~~~~~~~~ 📖 Textbook - Coming soon 😸🐙 GitHub - https://github.com/jeffheaton/app_generative_ai ▶️ Play List - https://www.youtube.com/watch?v=FBmUxUt__rM&list=PLjy4p-07OYzui0nVZzMgoLBeXjG9Oy3hi&ab_channel=JeffHeaton ~~~~~~~~~~~~~~~ CONNECT ~~~~~~~~~~~~~~~ 🖥️ Website: https://www.heatonresearch.com/ 🐦 Twitter - https://twitter.com/jeffheaton 😸🐙 GitHub - https://github.com/jeffheaton 📸 Instagram - https://www.instagram.com/jeffheatondotcom/ 🦾 Discord: https://discord.gg/3bjthYv ▶️ Subscribe: https://www.youtube.c
DeepCamp AI