SelfEvaluate Agent: Build a safe Agent with LangChain, LangGraph & LangSmith

Note & Mandatory: It's mandatory to both RSVP and fill this Google form https://forms.gle/Exhb1KvCd74M6mEcA if either is not done, security won't allow at the venue.

Note & Mandatory: Once you submit the Google Form, you will receive a confirmation email. Make sure that you keep it handy to show at the security entrance. If this email is not shown you will not be allowed entry.

What happens when an AI agent can read and modify its own source code?

In this session, we build and demo a minimal self-editing coding agent in Python that can inspect files, apply changes, run code, and validate outcomes in a controlled loop.
The architecture combines LangChain for model and tool orchestration, LangGraph for stateful workflow control, and LangSmith for tracing, observability, and evaluation. The system is model-agnostic by design, so you can swap LLM providers over time without redesigning the core agent workflow.
To keep autonomy practical and safe, every change goes through guardrails such as path restrictions, protected files, syntax validation, automatic backups, and run-verify-fix checks. We also cover multi-file consistency patterns so edits remain reliable as the codebase evolves.
By the end, you will have a clear blueprint for building autonomous coding agents.

Key takeaways:

How to design safe self-editing agent loops
How LangChain, LangGraph, and LangSmith work together
How to add observability and evaluation to agent behavior
How to keep the architecture model-agnostic for future LLM changes

Short blurb: A hands-on walkthrough of a safe self-editing coding agent built with LangChain, LangGraph, and LangSmith. Learn guardrails, evaluation patterns, and model-agnostic design so you can evolve your LLM choice without rebuilding your system.

Hackathon: After the break, we will also be having a Hackathon where users will be doing hands-on and building apps and showcasing them.