Agentic AI systems.

Multi-agent workflows that do real work instead of demoing well.

Start a project How I build it
  • LangGraph
  • LangChain
  • MCP
  • Python

The problem

An agent that works in a demo can still fail on step ten of a real task. It calls the wrong tool, loses the thread, or answers with confidence and nobody checks.

Here’s how I build it

How I build it.

5 steps from the first conversation to launch. The diagram fills in as each part gets built.

  1. Step 01

    Map the work

    I start with the job, not the model. We list the tasks the system must do, the data each one needs, and where a person has to approve an action.

  2. Step 02

    Build the tools

    Agents act through tools. Your APIs and business logic become tools with clear inputs, served from an MCP server so any agent or MCP client can call them.

  3. Step 03

    Wire the agents

    Each agent gets one job and only the tools it needs. LangGraph holds the state, routes between agents and retries a step that fails. Memory keeps the conversation together.

  4. Step 04

    Grade every answer

    Before users see it, a set of real questions runs through the system and each answer is graded against a rubric. A change that makes answers worse shows up as a lower score.

  5. Step 05

    Ship it and watch it

    It goes live with tracing on each call: which agent ran, which tools it used, how long it took. Problems show up on a dashboard before they reach a user.

What you get.

All of it, or only the part you need. Each piece can be scoped on its own.

Built with

  • LangGraph
  • LangChain
  • MCP
  • Python
  1. 01Agent graph design — state, routing, retries, human-in-the-loop
  2. 02Tool surface over your existing APIs and business logic
  3. 03An MCP server so any client can call those tools
  4. 04Rubric-graded evaluation before anything reaches users

Where I’ve done it.

Quoted from my CV, then the builds you can open and check for yourself.

  • Padelos

    Associate AI Engineer

    Jan 2026 — Present

    • Build LLM-powered agents and multi-agent workflows on the Padelos SaaS platform with LangChain and LangGraph, automating multi-step production processes.
    • Developed a Model Context Protocol (MCP) server that exposes platform data and business logic to AI agents through structured tool-calling interfaces.
    • Engineer and version system prompts for PadelGPT, the platform assistant, and built a quality-evaluation pipeline that grades responses against a rubric and generates structured Excel and CSV reports.

  • IR Solutions

    Junior AI Engineer

    Jul — Dec 2025

    • Built production agentic AI systems with n8n and engineered large-scale web scraping and data-ingestion pipelines.
    • Delivered end-to-end AI agents from design through deployment in a fast-paced team.
  • UH

    University of Haripur

    BS Computer Science — CGPA 3.79 / 4.0

    Sep 2021 — Jul 2025

    • Contributed to modelcontextprotocol/inspector, fixing a rendering bug where nested and optional array parameters fell back to raw JSON instead of structured form inputs.

Builds.

All projects

Questions clients ask.

Not covered here? Ask me directly.

Do I need several agents, or is one enough?

Often one agent with good tools is enough, and it is easier to run. I split the work into more agents only when the tasks really differ, such as reading data and changing it.

Can the agents use our existing systems?

Yes. Your APIs and business logic become tools the agents call. With an MCP server in front of them, other MCP clients can use the same tools later.

How do you know the answers are right?

A test set of real questions runs before each release, and every answer is graded against a rubric. You get the scores in a report, so quality is something you can track.

Which model do you use?

The one that fits the task and the budget. The agent graph doesn’t depend on one provider, so the model can change without rebuilding the system.

Other services.

Work with me

Have a project like this?

Tell me what you want to build, and I'll tell you how I would approach it.

Let’s talk