Production ML · Applied AI · Cloud systems

ROHANPATIL— Machine Learning Engineer portfolio

I build production ML systems that turn complex data into faster, more reliable decisions.

ML Engineer at PANTA, working end-to-end across model development, data pipelines, infrastructure, and product delivery.

Selected outcomes at SapientAI
Better prediction accuracy
25%
Production data processed
500GB+
Pipeline runtime reduction
40%
View selected workResume
Reinforcement Learning

AlphaZero for 2048

View case study
AI Agents

Autonomous Agents — LangGraph Due Diligence

View case study
Web App

Product Ingredient Analyzer

View case study

Selected work

Reinforcement Learning

AlphaZero for 2048

2048 has on the order of 10⁵² reachable boards and a random 2 or 4 after every move. Strong published agents still lean on hand-coded evaluation — monotonicity, smoothness, empty cells — which has to be rewritten if the rules change.

~28%
late games reach 2048
View case study
AI Agents

Autonomous Agents — LangGraph Due Diligence

Due diligence burns days pulling filings, news, financials, and prospect lists by hand. Insights arrive late, and the work does not scale.

Days → min
research cycle
View case study
Web App

Product Ingredient Analyzer

Packaging hides behind names like ALPHA-ISOMETHYL IONONE. The useful question is what they are and whether they are safe — without opening ten tabs.

1 photo
from label to brief
View case study
Web Development

Full-Stack Web Development

End-to-end sites on a serverless Cloudflare stack — Workers, R2, and D1 — spanning photography, hospitality, wellness, and media.

5 sites
across client verticals
View case study
Data Science

Sentiment Analysis Dashboard

Public opinion is a market signal, but comment threads are too noisy to read by hand.

View case study
NLP

News Aggregator & Clean News Feed

Clickbait, pop-ups, and jargon make it hard to get a reliable briefing from the open web.

View case study
Finance

SEC Filing Sentiment for Stock Evaluation

The useful signal in SEC filings is buried in technical jargon and volume.

View case study
Statistics

Statistical Workforce Productivity Study

The open question was which incentives actually change output — not which ones sound motivating.

View case study
Physics

Muon Mass via Monte Carlo Simulation

A 40,000-run Monte Carlo estimating muon mass at 100 ± 5 MeV, with automated parameter sweeps.

40k
Monte Carlo runs
View case study
Physics

Building a Microscope

A custom optical microscope from off-the-shelf parts, calibrated with Python image analysis.

View case study

Experience

Applied physics to production ML, cloud-native products, and data systems with measurable outcomes.

View resume

  1. Machine Learning Engineer

    2026 – Present

    Panta OS

    • Developing and deploying machine learning models for production systems.
    • Building scalable ML pipelines and cutting-edge AI infrastructure.
    model deployment
    Production
    ML ownership
    End-to-end
    cloud delivery
    Azure + AWS
    PythonML/DLAzureAWS

  2. Freelance Full-Stack Developer

    2024 – Present

    Self-Employed

    • Architecting high-performance web apps on the Cloudflare stack — Workers, Pages, D1, R2.
    • Next.js and Astro builds for clients across creative agencies, hospitality, and wellness.
    delivery model
    Edge-first
    Workers, Pages, D1, R2
    4 services
    Cloudflare WorkersD1 / R2Next.jsAstro

  3. Computational Analyst

    2022 – 2024

    SapientAI · Austin, TX

    • Built and deployed ML models for nuclear-fusion plasma simulations — 25% better prediction accuracy.
    • Engineered data pipelines processing 500GB+ datasets, cutting runtime by 40%.
    • Maintained and scaled 3 production applications for internal research teams.
    25%
    better prediction accuracy
    500GB+
    pipeline scale
    40%
    runtime reduction

  4. Teaching Assistant

    2019 – 2022

    UMass Amherst

    • Led weekly Physics & Astronomy lab sessions for 120+ undergraduate students.
    • Evaluated 1,200+ assignments and exams; detailed feedback lifted the class average by 15%.
    120+
    students taught
    1,200+
    assignments reviewed
    15%
    class-average lift
Earlier roles & education4 entries

  1. Masters in Data Science

    2025 – 2026

    COEP Technological University

    Advanced statistical modeling, machine learning, and big-data analytics applied to real-world problems.

  2. Marketing & Sponsorship Manager

    2019 – 2020

    CodeDay · Boston

    Sponsorships and marketing for a 200-attendee CS education event.

  3. BS in Applied Physics

    2018 – 2022

    UMass Amherst

    Major in Applied Physics, Minor in Resource Economics — plus a hard-earned respect for New England winters.

  4. Exchange Student

    2015 – 2016

    Rungsted Gymnasium · Denmark

    A year in Denmark: biking through blizzards, excessive pastries, and a working theory of hygge.

Capabilities

A focused set of technologies I can defend with production ownership or a working system—not a keyword inventory.

Production ML

Training, evaluating, and shipping models as reliable systems.

PythonPyTorchScikit-LearnHugging Face
See in AlphaZero training system

Applied AI & Agents

Source-aware agents, retrieval, orchestration, and model integration.

LangGraphLLM systemsRAGMCP
See in Autonomous due diligence

Data & Backend

Typed services and data layers that support ML products in production.

FastAPIPydanticPostgreSQLSQL
See in Ingredient analysis pipeline

Cloud & Delivery

Cloud-native deployment, observability, and edge-first application delivery.

AzureAWSCloudflare WorkersDocker
See in Cloudflare edge application

Discuss an ML/AI
opportunity.

Production ML, applied AI, data systems, or an unusually difficult technical problem.

Copied
--:--

ROHAN PATIL

Production ML engineer at PANTA, building reliable models, data pipelines, and AI products from prototype to production.

Work

Reinforcement Learning

AlphaZero for 2048

2048 has on the order of 10⁵² reachable boards and a random 2 or 4 after every move. Strong published agents still lean on hand-coded evaluation — monotonicity, smoothness, empty cells — which has to be rewritten if the rules change.

~28%
late games reach 2048
View case study
AI Agents

Autonomous Agents — LangGraph Due Diligence

Due diligence burns days pulling filings, news, financials, and prospect lists by hand. Insights arrive late, and the work does not scale.

Days → min
research cycle
View case study
Web App

Product Ingredient Analyzer

Packaging hides behind names like ALPHA-ISOMETHYL IONONE. The useful question is what they are and whether they are safe — without opening ten tabs.

1 photo
from label to brief
View case study
Web Development

Full-Stack Web Development

End-to-end sites on a serverless Cloudflare stack — Workers, R2, and D1 — spanning photography, hospitality, wellness, and media.

5 sites
across client verticals
View case study
Data Science

Sentiment Analysis Dashboard

Public opinion is a market signal, but comment threads are too noisy to read by hand.

View case study
NLP

News Aggregator & Clean News Feed

Clickbait, pop-ups, and jargon make it hard to get a reliable briefing from the open web.

View case study
Finance

SEC Filing Sentiment for Stock Evaluation

The useful signal in SEC filings is buried in technical jargon and volume.

View case study
Statistics

Statistical Workforce Productivity Study

The open question was which incentives actually change output — not which ones sound motivating.

View case study
Physics

Muon Mass via Monte Carlo Simulation

A 40,000-run Monte Carlo estimating muon mass at 100 ± 5 MeV, with automated parameter sweeps.

40k
Monte Carlo runs
View case study
Physics

Building a Microscope

A custom optical microscope from off-the-shelf parts, calibrated with Python image analysis.

View case study

Experience

Resume
  1. Machine Learning EngineerPanta OS · 2026 – Present
    • Developing and deploying machine learning models for production systems.
    • Building scalable ML pipelines and cutting-edge AI infrastructure.
    PythonML/DLAzureAWS
  2. Freelance Full-Stack DeveloperSelf-Employed · 2024 – Present
    • Architecting high-performance web apps on the Cloudflare stack — Workers, Pages, D1, R2.
    • Next.js and Astro builds for clients across creative agencies, hospitality, and wellness.
    Cloudflare WorkersD1 / R2Next.jsAstro
  3. Computational AnalystSapientAI · 2022 – 2024
    • Built and deployed ML models for nuclear-fusion plasma simulations — 25% better prediction accuracy.
    • Engineered data pipelines processing 500GB+ datasets, cutting runtime by 40%.
    • Maintained and scaled 3 production applications for internal research teams.
  4. Teaching AssistantUMass Amherst · 2019 – 2022
    • Led weekly Physics & Astronomy lab sessions for 120+ undergraduate students.
    • Evaluated 1,200+ assignments and exams; detailed feedback lifted the class average by 15%.
Earlier roles & education4
  1. Masters in Data ScienceCOEP Technological University · 2025 – 2026

    Advanced statistical modeling, machine learning, and big-data analytics applied to real-world problems.

  2. Marketing & Sponsorship ManagerCodeDay · 2019 – 2020

    Sponsorships and marketing for a 200-attendee CS education event.

  3. BS in Applied PhysicsUMass Amherst · 2018 – 2022

    Major in Applied Physics, Minor in Resource Economics — plus a hard-earned respect for New England winters.

  4. Exchange StudentRungsted Gymnasium · 2015 – 2016

    A year in Denmark: biking through blizzards, excessive pastries, and a working theory of hygge.

Capabilities

Production ML

Training, evaluating, and shipping models as reliable systems.

PythonPyTorchScikit-LearnHugging Face
See the proof
Applied AI & Agents

Source-aware agents, retrieval, orchestration, and model integration.

LangGraphLLM systemsRAGMCP
See the proof
Data & Backend

Typed services and data layers that support ML products in production.

FastAPIPydanticPostgreSQLSQL
See the proof
Cloud & Delivery

Cloud-native deployment, observability, and edge-first application delivery.

AzureAWSCloudflare WorkersDocker
See the proof

Contact

Discuss an ML/AI
opportunity.

Copied
--:--