01 / about
I care more about what breaks than what demos well.
I'm an Applied Machine Learning graduate from the University of Maryland, and most of what I build now sits inside agentic AI — LangGraph pipelines, ReAct loops, and retrieval-augmented systems that can plan and act instead of just answering a single prompt. That interest didn't come out of nowhere: before agents, I spent time in the more traditional side of ML, designing and tuning models from scratch, including a residual neural network built for an industrial forecasting problem at CNPC. Agent orchestration feels like a natural next step from that work — the same core problem of getting a system to make good decisions on incomplete information, just at a higher level of abstraction.
Alongside that, I've spent the past year as a Graduate Teaching Assistant at UMD, supporting 300+ students across data science and machine learning coursework. I also spend a fair amount of free time on competitive programming, particularly graph problems, which turned out to be surprisingly good training for thinking about agent orchestration too.
02 / education
Two campuses, one throughline.
2024 — 2026
University of Maryland
M.S. in Applied Machine Learning
College Park, MD
2020 — 2024
Vellore Institute of Technology (VIT)
B.E. in Artificial Intelligence
Vellore, India
03 / skills
What I actually reach for.
Agentic AI & LLM Systems
Languages
ML & Data
Infra & Tooling
Tools & Platforms
Data & Retrieval
04 / experience
Where the work happened.
Machine Learning Intern
CNPC USA · Houston, TX
- Developed deep learning systems for Rate of Penetration (ROP) prediction
- Improved R² from 0.66 → 0.75 over production XGBoost baselines
- Designed reproducible ML preprocessing and deployment pipelines
- Applied Bayesian Optimization for hyperparameter tuning
- Deployed models in production drilling operations
JUN 2026 — SEPT 2026
Graduate Teaching Assistant
University of Maryland · College Park, MD
- Led hands-on coding labs and mentored 300+ students
- Graded and provided feedback on 6,000+ data science assignments
- Created supplementary educational materials for ML courses
JAN 2025 — MAY 2026
05 / portfolio
Production grade AI, one pipeline at a time
April 2026
An autonomous agent that resolves GitHub issues end to end.
more detail +less detail −
A ReAct-style agent built on LangGraph that reads an open GitHub issue, reasons about the fix, edits the codebase, and opens a PR. Demonstrated against a live Flask repository, with Claude as the reasoning engine, Docker for sandboxed execution, and LangSmith for tracing every step of the agent's decisions.
March 2026
A multi-stage LLM pipeline that won the GDG Hackathon.
more detail +less detail −
A job search platform that chains several LLM stages together to parse, match, and rank opportunities against a candidate's profile. Built on Next.js and TypeScript with Firebase for auth and data, deployed on Vercel.
December 2025
A RAG assistant, and a real lesson in training-serving skew.
more detail +less detail −
A retrieval-augmented question answering system over dental domain documents, built on FAISS for vector search. The interesting part wasn't the happy path — it was a training-serving skew bug in production that taught me to take evaluation parity between offline and online pipelines a lot more seriously.
November 2025
A multi-agent trip planner built on Google's Agent Development Kit.
more detail +less detail −
A planning agent that breaks a travel request into sub-tasks — flights, lodging, itinerary sequencing — and coordinates specialized sub-agents to fill each one in, built using Google ADK.
07 / contact
Reach out — I answer fast.
Amaan Mohammed — built with Next.js & Tailwind
© 2026