Sellable
An agent-ready storefront where AI buyers can find products and negotiate within seller-set rules. Purchases pass through consent checks, spending limits, and Razorpay test payments, with each step recorded for review.
I build AI systems, agents,
and automations. [more]
making ai do more
than just yap.
Mahir Malik
MahirMalik.
AI Engineer
Sitemahirmalik.in
AI SystemsAgentsAutomation
I’m an AI engineer. I build AI agents, agentic systems, and automations that take care of real tasks. I do applied AI work too, adding language models to apps and software and making them work with the rest of the system.
I train machine learning models and build the systems around them. I also work on fine-tuning, deep learning, and reinforcement learning. I like taking a model from an experiment to something that runs as part of a product.
I build full-stack web apps with AI features, as well as apps in Rust. I work across the software stack, from the frontend and backend to APIs and databases. I also deploy and run my apps in the cloud, on virtual machines, and in Kubernetes clusters.
Confluence
An interactive way to see machine learning at work. Compare 38 scikit-learn algorithms, explore 25 datasets, and step through how models learn with decision-boundary views, loss curves, and a tree builder.
Helion
An agent harness for your terminal and desktop. Watch runs unfold, steer them mid-flight, and approve tool calls as they happen.
- TypeScript
- Tauri
- Ink
- React
Latest
Weight-Space Programming
A plain-language guide to GPT-6 Astra’s reported Qwen3.5-0.8B weight edit: the Python tools it used, the reasoning loop it targeted, the benchmark results, and the limits of the claim.
Jev Explained
Jev by TypeSafe AI is a System One model for fast typed decisions: how its Choice, Score and Noul primitives work, what it costs, and where it fits beside an LLM.
Context vs Loop vs Harness Engineering
How context engineering, harness engineering, and loop engineering differ, where they overlap, and how to build a simple production-style agent system.
I made a visualization tool for ML algorithms
Confluence is a playground for learning ML by watching it work: 38 real scikit-learn algorithms, decision boundaries you can poke at, and training you can scrub through frame by frame.
Implementations
Because reading the paper wasn't enough.
I take interesting papers, ideas, and algorithms and actually build them from scratch.
Then I test them with different data, compare the results, and write down what I learn. It's a hands-on way to see how an idea works beyond the page.
papers.mahirmalik.inWhat I build with.
A practical toolkit for taking ideas from a first experiment to something people can use.
- Python
- Rust
- Tailwind CSS
- SQLAlchemy
- GraphQL
- Uvicorn
- SciPy
- Jupyter
- PyTorch
- Transformers
- Ollama
- LangGraph
- PydanticAI
- Google ADK
- Codex
- Orca
- MySQL
- n8n
- AWS
- Kubernetes
- Railway
- Terraform
- OpenTelemetry
- Langfuse
- VS Code
- Bun
- MLflow
- TypeScript
- React
- FastAPI
- Alembic
- gRPC
- NumPy
- Matplotlib
- PyArrow
- scikit-learn
- LangChain
- vLLM
- CrewAI
- MCP
- OpenCode
- Hermes
- Herder
- Redis
- Zapier
- Oracle Cloud
- Cloudflare
- Fly.io
- Nginx
- Prometheus
- Git
- Postman
- pnpm
- SQL
- Next.js
- Pydantic
- WebSockets
- Celery
- Pandas
- Plotly
- Polars
- Hugging Face
- LlamaIndex
- LiteLLM
- AutoGen
- OpenAI SDK
- Claude Code
- Pi
- PostgreSQL
- Supabase
- Temporal
- Docker
- Vercel
- Render
- GitHub Actions
- Grafana
- GitHub
- Linux
- uv
GitHub Activity
Contribution graph and recent commits.
A GitHub-style overview of public contributions in 2026, followed by recent public activity and commit messages from @mahirmlk.
Loading
View Profile 0 contributions in 2026
Recent Public Activity
Latest visible actions pulled from the GitHub public events feed.
No recent public activity.
No recent public push commits are available right now, but the activity feed above is live.