About the role
We're building a chat-based assistant that lets advertisers optimize their Meta campaigns in plain language. Users ask open-ended questions like /"I want more leads,/" /"my leads are coming from too far away,/" or /"my leads aren't answering./" The system has to work out what's actually wrong, pull the right campaign data, and recommend or take action.
You'll design and build the agentic core of this product. It's an active, in-development project, so you'll be joining a live codebase, not starting from a blank page.
What you'll do
Design and build a multi-step agent that turns vague user goals into diagnosis and action on Meta campaigns
Implement tool/function calling against our FastAPI backend and the Meta Marketing API
Build agent workflows with LangGraph (state, branching, retries, human-in-the-loop checkpoints)
Enforce structured outputs (JSON schema / Pydantic) so agent actions are protected and predictable
Add guardrails so the agent never takes a costly campaign action without validation
Set up evaluations and tracing (e.g. LangSmith or Langfuse)



to measure accuracy, latency, and cost per conversation
Containerize and deploy services on AWS (EKS, EC2) with Docker
Work closely with the frontend team (Next.js) on streaming chat responses and agent state in the UI
Must have
2+ years of hands-on Python development, including async code and API design with FastAPI
At least one agentic system you have built and shipped (not just a tutorial or wrapper around a single prompt)
Practical experience with LLM tool/function calling and multi-step agent patterns (ReAct, planner–executor, or graph-based workflows)
Working knowledge of LangGraph or an equivalent orchestration framework
Experience evaluating LLM outputs: test sets, failure analysis, prompt iteration based on results
Docker, AWS (EKS, EC2), Linux servers and SSH access
Git and GitHub workflows (branches, PRs, code reviews)
Explicit written and spoken communication in En

📌 Ai Engineer: Agentic Systems Panchkula
🏢 Blynk Ads
📍 Panchkula