S Naga Surya AI / Backend Engineer
Resume

Case study · 2025

Eco Online

Agent-driven software delivery platform

Refines Jira tickets, hands implementation to AI coding agents, tracks pull requests through deployment, and runs live acceptance tests.

Ingesttickets over MCPticketRefineacceptance criteriatasksBuildAI coding agentsShipPR, merge, deployafter deployTestlive tests by ITAFMCP integrations I builtno manual hand-offs Ingesttickets over MCPticketRefineacceptance criteriatasksBuildAI coding agentsShipPR, merge, deployafter deployTestlive tests by ITAFMCPintegrations Ibuiltno manualhand-offs
How Eco Online works. The coloured notes are my annotations.

What it does

Eco Online runs the software delivery lifecycle with agents. It combines PO Agent for planning and ITAF for testing: a ticket’s scope and acceptance criteria are refined, implementation goes to AI coding agents, pull requests are tracked through merge and deployment, and ITAF verifies every acceptance criterion against the live build.

I owned the FastAPI and LangGraph backend, including the MCP integrations and the Anthropic Claude and OpenAI tool-calling loops.

How it works

  1. Ingest

    MCP integrations fetch Jira tickets, acceptance criteria, repository context, pull-request state and deployment information.

  2. Refine

    Workflows tighten ticket scope and acceptance criteria before any implementation starts.

  3. Build

    Implementation tasks are handed to AI coding agents.

  4. Ship

    The platform opens GitHub pull requests and tracks merge and deployment state, with no manual workflow coordination.

  5. Test

    After deployment, ITAF generates and runs live tests mapped to each ticket’s acceptance criteria.

What I built

  • Owned the FastAPI and LangGraph backend for a client platform combining PO Agent and ITAF into an agent-monitored delivery lifecycle.
  • Built MCP integrations to fetch Jira tickets, acceptance criteria, repository context, pull-request state and deployment information.
  • Created workflows that refine ticket scope and acceptance criteria before passing implementation tasks to AI coding agents.
  • Orchestrated GitHub pull-request creation and monitored merge and deployment state without requiring manual workflow coordination.
  • Triggered ITAF after deployment to generate and execute live tests mapped directly to each ticket’s acceptance criteria.
  • Integrated Anthropic Claude and OpenAI using tool calling, multi-turn loops and prompt caching; stored workflow data in Supabase.