BUILDER · AGENTIC AI

Give me the problem. I ship the system.

17, coding since Class 8, running about six tabs in my head at once, and still shipping real things: production apps and AI agents for a startup, freelance clients, and 3,700+ people who installed something I built. Once I start, I finish. No matter what.

See the deployments
Muhammad Marjan Ahmed

Muhammad Marjan Ahmed

Builder — Agentic AI Systems

Base
Karachi, Pakistan
Local time
--:--:-- PKT
Focus
Agentic AI systems
Status
Open to co-build

Proof

0+PyPI downloadsGemini Starter Agent, open source0+Public reposshipped code on GitHub
0Client sitesshipped end to end at Reintechs
0Hackathonsfinalist, top 10, 2nd place, since 2024
0Agecoding since Class 8
001METHOD

How I work when I'm dropped into a problem

Five habits, each with a receipt. A principle without evidence is just a slogan.

  1. 01

    Start from friction

    I pick problems that have real friction behind them, not the ones that look good in a portfolio, then figure out how to build the solution.

    Receipt:DocSamajh AI

  2. 02

    Constraints are the brief

    A three-hour build with a live clock is what pushed me to read the OpenAI Agents SDK source closely enough to contribute a feature back, not just use it.

    Receipt:3-hour hackathon → 2nd place

  3. 03

    Own it end to end

    5–7 client sites, solo, from CMS to checkout to the write-up the client actually reads. No handoffs for the parts that aren't fun.

    Receipt:5–7 client sites, solo

  4. 04

    Think in systems

    Three services, three platforms, one person, because a real system is judged on what happens when a piece of it fails, not on the demo.

    Receipt:3 services, 3 platforms

  5. 05

    Make it checkable

    3,723+ PyPI downloads, not "used by developers." If a number is on this site, it's a number you can go verify.

    Receipt:3,723 downloads

  6. Everything above is checkable. That's on purpose. I'd rather you find the gaps yourself than take my word for it.

002DEPLOYMENTS

What actually shipped

Five systems, end to end. The friction that started them, the constraint I designed around, and what moved because of it.

  1. Friction
  2. Constraint
  3. System
  4. Impact
Live2025 — presentSelf-initiated

Physical AI & Humanoid Robotics

AI-native learning platform

Friction

Docs-style course content with no way to ask a question and get an answer grounded in the actual curriculum.

3services, 3 platforms
0hosting budget
5curriculum domains
Constraint

Zero hosting budget, one person building three separate services (docs, auth, RAG) on three different platforms.

System

A Docusaurus frontend on GitHub Pages, an Express + Better Auth service, and a FastAPI RAG backend on Railway using Cohere embeddings in Qdrant. Three services, three platforms, held together by one person.

Physical AI & Humanoid Robotics
Architecture
  1. Course frontend

    Docusaurus · Pages

  2. Auth service

    Express · Better Auth

  3. RAG API

    FastAPI · UV · Railway

  4. Embeddings

    Cohere

  5. Vector store

    Qdrant

Stack
  • Docusaurus
  • FastAPI
  • UV
  • Cohere
  • Qdrant
  • Express
  • Better Auth
  • Neon Postgres
LiveSep 2025Self-initiated

Gemini Starter Agent

Open source · PyPI

Friction

Every new agent project started with the same thirty minutes of provider boilerplate before you could write a single line of actual agent logic.

3,723+PyPI downloads
<60sto a running agent
160+public repos behind it
ShippedDec 2025Hackathon build

DocSamajh AI

AI Financial Hackathon · LandingAI

Friction

Invoice-to-PO reconciliation was manual, line by line, the actual bottleneck the hackathon brief was built around.

85–90%match accuracy
~70%manual effort removed
Shipped2025Hackathon build

AI Employee Vault

GIAIC Hackathon 0

Friction

Signals worth acting on (email, WhatsApp, filesystem changes) scattered across channels with no single place to review and approve action.

100%writes gated on approval
3channels watched
1readable audit trail
ShippedJan–Feb 2025Hackathon build

Hekto Marketplace

Hackathon Sprint · GIAIC

Featured — GIAIC Demo Day

Friction

A hackathon brief for a full furniture e-commerce platform (catalog, cart, checkout, shipping) with nothing built yet.

6 dayszero to deployed
18QA test cases logged
8/8checkout flow tests passed
003CAPABILITIES

How I operate

Grouped by the problem each one solves, not by language, so it reads the way the work actually happens.

  • 01

    Ship the interface

    The layer a user actually touches: fast, accessible, and built with the same stack across every project so nothing is a one-off.

    • TypeScript
    • Next.js
    • React
    • Tailwind CSS
    • shadcn/ui
    • Vercel
  • 02

    Reason over data

    Agents, retrieval, and multi-agent orchestration. The part of the system that decides, not just responds.

    • OpenAI Agents SDK
    • Model Context Protocol
    • RAG
    • Cohere
    • Qdrant
    • Multi-agent orchestration
    • Claude Code
  • 03

    Serve it reliably

    APIs, auth, and storage that keep running after the demo is over.

    • Python
    • FastAPI
    • Node.js
    • Express
    • Supabase
    • Neon Postgres
    • Sanity CMS
  • 04

    Run it in production

    Containers, CI, and hosting chosen for zero budget and real uptime, not for the resume line.

    • Docker
    • Git
    • GitHub Actions
    • Railway
    • GitHub Pages
    • FastAPI Cloud
004LOG

Everything that shipped

Two years, one timeline, newest first.

  1. Attended

    KhiNext Tech Summit, Karachi

    "AI, Commerce & The Future Consumer" panel.

    At the KhiNext tech summit in Karachi
  2. Joined

    Naxasware, Full Stack Developer

    Shipping production features with the founding team.

  3. Joined

    Microsoft Student Ambassadors, Karachi

    Event management team.

  4. Attended

    DevFest Karachi, GDG Kolachi's 10th anniversary

    Community event at NASTP.

    At DevFest Karachi, GDG Kolachi's 10th anniversary event
  5. Shipped

    AI Financial Hackathon, Submitted

    LandingAI, DocSamajh AI.

  6. Won

    Agentic AI Hackathon, Finalist

    IBM watsonx, lablab.ai.

  7. Won

    Cloudways x DigitalOcean Hacktoberfest, 2nd Place

    Three-hour build. Contributed a feature to OpenAI Agents SDK.

    Won · Oct 2025 · Karachi
    Team announced 2nd place at Hacktoberfest, with judge Aaqib Sayed
    Field note

    Led the team. Two hours in, we had nothing working. The SDK didn't have what the brief needed. We found the gap, built around it, and didn't have a working demo until the last hour. 2nd place, standing next to Aaqib Sayed, the judge who called it.

  8. Shipped

    NASA Space Apps, Kalkia

    Weather probability over 44 years of NASA data.

    Shipped · Oct 2025 · Karachi
    Presenting Kalkia in the NASA Space Apps Karachi Discord
    Field note

    First time being an ambassador for anything. I made a Google form, posted it around school, got 16 responses. Then built Kalkia for the challenge myself: weather probability from 44 years of NASA data.

  9. Published

    Gemini Starter Agent, PyPI

    3,723+ downloads.

  10. Joined

    Reintechs, Web Developer (Freelance)

    5 to 7 production client sites over five months.

  11. Shipped

    National Agentic AI Hackathon, Innovista Indus

    First hackathon. EdTech in Urdu, Sindhi, Punjabi, Pashto.

    Shipped · 2024 · Karachi
    At Innovista Indus for the National Agentic AI Hackathon
    Field note

    My first hackathon. The idea was right, AI responses in Urdu, Sindhi, Punjabi, and Pashto, but I was nowhere near as fast or as organised as I am now. Everything since has been catching up to that idea.

  12. Won

    Google Build with AI, Top 10

    GDG Kolachi, Hunar Bazaar.

    Won · Jun 2024 · Karachi
    Team presenting Hunar Bazaar at Build with AI
    Field note

    Top 10 out of the field, GDG Kolachi. We built Hunar Bazaar, a freelancing platform for rural Pakistani women, with AI-generated gig descriptions.

  13. Started

    GIAIC, Agentic AI & Cloud Engineering

    Alongside matriculation.

005ROADMAP

First on my roadmap: an AI-native company on a Forward-Deployed-Engineer Agent Factory (FDE-AF) model, agents shipped into real companies, not demos.

Partnering with ambitious startups now is where that starts.

Not a receipt like the rest of this page — this is direction.

006Contact

Bring me the problem

If you're an AI startup, anywhere, looking for a long-term partner to own the system through to scale, not another hire, that's what I'm looking for. Reach out via email or LinkedIn.

Select Language