just an ai engineer with a thing for understanding what makes systems work.

working across ml, llms, and ai agents, and full-stack web apps with a lot of curiosity around how models learn, how agents reason, and how all of that fits into software.

because integrating models was not enough, so diving deep into deep learning, and rl stuff to take datasets and turn them into working models. also looking forward to do open-source contributions in relevant field and build some projects too.

portfolio · linkedin · x · kaggle

Sellable

Agentic commerce, with actual rules around the money. AI agents can discover products, negotiate prices, request consent, and complete purchases. Spending limits, policy checks, payment verification, and audit trails don't get handed over to the LLM.

Python FastAPI LangGraph Next.js Supabase PostgreSQL Razorpay

Helion · in progress

An agent harness for the terminal and desktop. Run agents, follow their execution, steer them mid-run, and approve tool calls without losing sight of what's happening.

TypeScript Tauri React Ink

• agentic systems: tool use, orchestration, memory, evaluation, and harness engineering.
• distributed systems: concurrency, state, communication, failure handling, and system design.
• reinforcement learning: policies, reward design, exploration, and decision-making.
• ai security: trust boundaries, prompt injection, tool permissions, and what happens when an agent shouldn't be trusted.
• applied ai: model behavior, inference, retrieval, and getting ML out of notebooks and into actual software.

I like understanding the internals, testing assumptions, and building small implementations to see where the theory meets reality.

the tools i reach for across ai, backend, and infrastructure.

Portfolio · GitHub · LinkedIn · X · Kaggle · Paper implementations

For work, collaborations, or an interesting technical work: mahirmalikx@gmail.com

build it, break it, understand it, repeat.