Frameworks and services I keep in my workbench.

PyTorch

My primary framework for research. Every paper I've published, every model I've trained — including MEDiC, MaskDistill-PyTorch, and Cross-Scale MAE — runs on PyTorch.

HuggingFace

Pretrained backbones, datasets, and where I publish my own trained models and model cards. The first place I look when starting a new experiment.

CUDA

GPU compute for training at scale. I live in nvidia-smi during research sprints and long training runs.

Jupyter

Where every experiment starts. Quick prototyping, plotting, and sanity-checking model outputs before they graduate into scripts.

Weights & Biases

Experiment tracking, hyperparameter sweeps, and model checkpoints for every training run. The loss curves that keep me honest.

NumPy

The backbone of every ML workflow. Data preprocessing, tensor math, and quick sanity checks before things become a Tensor.

Pandas

Daily driver for data exploration, cleaning, and feature engineering across research and production projects.

scikit-learn

Classical ML baselines and quick prototyping before I reach for deep learning. Still the cleanest API in ML.

Python

My primary language for research and backend work. Also how I've shipped 7 PyPI packages over the years.

FastAPI

Go-to for Python backends. Powers the APIs behind FleetSmart.ai and most of my side projects. Typed, fast, async by default.

PostgreSQL

The default database for every web product I build. Reliable, extensible, and I trust it with production data.

Supabase

Postgres + auth + storage + realtime in one package. I used it to ship ShiftMD fast without babysitting infra.

Redis

In-memory cache, queue, and rate limiter for when Postgres needs a little help. Used everywhere a millisecond matters.

Pydantic

Runtime data validation and type-safe models in Python. FastAPI's best friend and the thing that keeps my APIs honest.

Poetry

Python dependency and packaging management. Every PyPI package I publish goes through Poetry.

TypeScript

Every line of frontend code I write is typed. Catches bugs at compile time so I don't ship them to production.

Next.js

Full-stack React framework behind FleetSmart.ai, ShiftMD, XpensAI, and Soma. App Router, server components, the whole thing.

React

UI component library I reach for whenever I need interactive client-side work.

Tailwind CSS

Utility-first styling for every frontend I build. I haven't written a stylesheet from scratch in years.

Astro

The framework powering this very site. Static by default, React islands where I need them, fast builds, minimal JS.

Framer Motion

Production-grade animations for React. Behind every hover, scroll reveal, and sparkle trail you see on this portfolio.

shadcn/ui

Copy-paste, own-the-code React components. The UI primitives I reach for first when I start a new product.

AWS

S3, Lambda, ECS, and RDS for production workloads. My daily infrastructure for the day job and side projects.

GCP

Compute and storage for FleetSmart.ai's backend. Cloud Run, Cloud Storage, and Cloud SQL for the heavy lifting.

Docker

Containerization for reproducible ML experiments, service deployments, and everything in between.

Kubernetes

Orchestration when a single container isn't enough. Used for scaling inference services and batch training jobs.

Cloudflare

CDN, DNS, Workers, and R2 storage. The edge layer sitting in front of most of my production workloads.

Vercel

Where my Next.js apps and this portfolio get deployed. Preview URLs, edge functions, and zero-config deploys.

GitHub Actions

CI/CD for all my repos. Tests, linting, builds, and deployments on every push.

CircleCI

CI/CD for my 7 PyPI packages. Runs the test matrix and publishes releases on version tags.

VS Code

Primary editor for everything from research notebooks to production code. Extensions, debugging, git integration — the whole kit.

Cursor

VS Code with an AI pair programmer built in. I reach for it when I want Claude writing alongside me.

Git

Version control for all 59 of my GitHub repos. Obviously.

Obsidian

Where I take notes on papers I read, meetings, and half-baked ideas that might become something.

Linear

Issue tracking and project planning for everything I ship. Keyboard-first, fast, and the opposite of Jira.

Figma

Design, wireframes, and the occasional quick mockup when I'm brainstorming a new feature or a layout.