Skip to content

Data & AI Engineer

ROHAN   PATIL – Data & AI Professional Portfolio

about

Data & AI Engineer with 3+ years of experience building analytics and AI products. Specialized in Python, ML/DL, Generative AI and LLMs, with end-to-end cloud deployment experience on Azure, AWS, Cloudflare.

I also do front end occasionally :)

Email
Copied
Based In
Pune, MH · --:--
Elsewhere
Work

Projects

Reinforcement Learning

2048 Master — AlphaZero RL

Reinforcement Learning

2048 Master — AlphaZero RL

The Problem

2048 is highly stochastic — tile spawns are random and the state space is enormous. Prior bots either lean on hand-tuned heuristics (Expectimax) or shallow planning (Deep Q-Learning), both leaving wins on the table.

AlphaZero-style agent that masters 2048 through pure self-play — no human heuristics. A custom 1D bitboard engine simulates thousands of games per second. A ResNet policy/value network is paired with Monte Carlo Tree Search for lookahead, trained via concurrent self-play, 8-way board-symmetry augmentation, and experience replay. Read the full report.

Stack

Python PyTorch AlphaZero MCTS ResNet Bitboards Reinforcement Learning
View Project
Web Development

Full-Stack Web Development

Web Development

Full-Stack Web Development

The Problem

Professional online presence without the technical know how.

I deliver end-to-end web solutions, from design to deployment. Here is a photography portfolio, with a custom CMS to manage all site content. This project runs on a serverless Cloudflare stack (Workers, R2, D1). Other projects include sites for a local bakery, a breath coach, a media studio, and a fine-dining restaurant.

Stack

Next.js React TypeScript Cloudflare Stack SQLite Tailwind CSS Framer Motion PostgreSQL
Web App

Product Ingredient Analyzer

Web App

Product Ingredient Analyzer

The Problem

Product marketing and packaging are very misleading. There are too many obscure ingredients like 'ALPHA-ISOMETHYL IONONE' or 'SOY LECITHIN, MONO + DIGLYCERIDES, CARAGEENAN'. I want to know what they are and if they are safe.

Snap or drop a product label to analyze it instantly. The app uses Gemini Vision for OCR, classifies ingredients, flags risky additives, and leverages Perplexity-powered LLM search for real-time insights—all running on Cloudflare edge.

Stack

Gemini Vision React TypeScript Cloudflare Workers OpenAI OCR
Live Demo GitHub
Data Science

Sentiment Analysis Dashboard

Data Science

Sentiment Analysis Dashboard

The Problem

Evaluating public opinion organically is a great way to understand the market. This can be applied to any product or service.

Interactive dashboard that analyzes YouTube comments—tracks sentiment with VADER and visualizes trends, keyword spikes, and word clouds.

Stack

Python React TypeScript SQLite Cloudflare Workers VADER YouTube API
Live Demo GitHub
NLP

News Aggregator & Clean News Feed

NLP

News Aggregator & Clean News Feed

The Problem

Clickbait news articles are worse than ever. Closing pop-ups, accepting cookies and filtering through all the jargon is hardly what our ancestors expected us to be doing.

Flask service scraping multi‑source articles, applying transformer stance & credibility models, and surfacing reliable headlines with short summaries.

Stack

Python Flask HuggingFace Transformers BeautifulSoup MongoDB
GitHub
Finance

SEC Filing Sentiment for Stock Evaluation

Finance

SEC Filing Sentiment for Stock Evaluation

The Problem

Reading through all the SEC filings is difficult. We need to be able to dissect the technical jargon and analyze the sentiment of the company.

Pipeline pulling 8‑K/10‑K filings, cleaning text, running FinBERT sentiment, and exporting buy/sell signals to Excel.

Stack

Python BeautifulSoup FinBERT SEC API Excel
GitHub
Statistics

Statistical Workforce Productivity Study

Statistics

Statistical Workforce Productivity Study

The Problem

Workers are not motivated to do their best work. We need to know how to motivate them.

MTurk experiment measuring how incentives & demographics shift output using T‑tests and regression.

Stack

Excel JMP MTurk
Physics

Muon Mass via Monte Carlo Simulation

Physics

Muon Mass via Monte Carlo Simulation

40k‑run Monte Carlo estimating muon mass at 100 ± 5 MeV with automated parameter sweeps.

Stack

Python NumPy SciPy Matplotlib
GitHub
Physics

Building a Microscope

Physics

Building a Microscope

Custom optical microscope built from off‑the‑shelf parts, calibrated via Python image analysis.

Stack

Python OpenCV CAD
Confidential
? Stealth project — under wraps
Confidential

Mystery Project

I am currently working on a stealth project that leverages advanced AI agents to automate complex software engineering tasks. This system is designed to autonomously parse requirements, plan implementation steps, and generate production-ready code with minimal human oversight.

Stack

LLMs Autonomous Agents Rust TypeScript
Career

Experience

2026

Machine Learning Engineer

Panta OS

Developing and deploying machine learning models for production systems. Building scalable ML pipelines and working on cutting-edge AI infrastructure.

2024

Freelance Full-Stack Developer

Self-Employed

Architecting and deploying high-performance web applications using the Cloudflare stack (Workers, Pages, D1, R2). Specializing in Next.js and Astro for clients in various industries including creative agencies, hospitality, and wellness.

Masters in Data Science

COEP Technological University

Specializing in advanced statistical modeling, machine learning, and big data analytics. Applying computational techniques to solve complex real-world data problems.

2022

Computational Analyst

SapientAI, Austin, TX

Built and deployed ML models for nuclear fusion plasma simulations, achieving 25% improvement in prediction accuracy. Engineered data pipelines processing 500GB+ datasets, reducing runtime by 40%. Maintained and scaled 3 production applications serving internal research teams.

2018

BS in Applied Physics

UMass Amherst

Major in Applied Physics, Minor in Resource Economics. Spent 4 years calculating trajectories, analyzing economic models, and discovering that New England winters are no joke.

Teaching Assistant

UMass Amherst

Led weekly lab sessions for 120+ undergraduate students in Physics and Astronomy. Evaluated 1,200+ assignments and exams, providing detailed feedback that improved the class average by 15%.

Marketing & Sponsorship Manager

CodeDay, Boston

Negotiated and secured $10K+ in sponsorships from 5 tech companies for a 200-attendee CS education event. Directed a cross-functional team of 4 and developed marketing campaigns.

2015

Exchange Student

Rungsted Gymnasium, Denmark

Lived in Denmark for a year. Mastered the art of riding a bike in a blizzard, eating excessive amounts of pastries, and understanding 'Hygge'.

1999

Hello World

Earth v1.0

Successfully deployed to production. Zero bugs reported at launch, Y2K immunity confirmed.

Toolkit

Skills

Languages

Python JavaScript TypeScript SQL HTML/CSS

Frameworks & Libraries

React Next.js Astro Tailwind CSS Flask Three.js

Tools & Platforms

Cloudflare Workers Netlify Vercel AWS GCP Docker Figma PostgreSQL SQLite PowerBI Tableau

Data Science

NumPy Pandas Scikit-Learn NLP Matplotlib TensorFlow PyTorch Keras Scipy