Rohan Patnaik

Now: Lead ML Engineer @ Altimetrik

Hi, I'm Rohan.
I build ML systems and the tools around them.

Lead ML engineer in India building data and ML orchestration on Databricks, working mostly in Python, Go and TypeScript. Outside work I ship developer tools like Codebloom, and send fixes upstream to the open-source projects I depend on.

Open to remote roles, or on-site in San Francisco, Hyderabad, Seattle, New York, Amsterdam, Copenhagen, Frankfurt and Munich.

See What I've Shipped Follow on X
Merged PRs by languageHover or tap a node

Things Shipped

  1. CodebloomSee any codebase bloom into a mind map, with call arrows across files. On VS Code, Open VSX and JetBrains. IDE Extension
  2. OrbitA better Alt+Tab for Omarchy: macOS icons, Windows 7 stacks or a Windows 11 grid. Omarchy
  3. subpoolYour Claude, ChatGPT and Gemini subscriptions as one fast API, with a built-in dashboard. Rust
  4. Pasture ImpostersAmong Us on a Pokémon-style farm: make your avatar and catch the imposter. Game
  5. OpenCalendlyOpen-source Calendly: share a link, take 1:1 or team bookings, synced to your calendar. Web App
  6. ChromarchyA local-first Photoshop alternative for Omarchy and Arch Linux. In Progress
  7. gogcli-plusDrive reports, local-to-Drive sync and contact deduplication on top of gogcli. CLI
  8. PaperloopDrive through Ilya Sutskever's must-read AI papers in a Minecraft-style world. Web App
  9. SolitaireFive classic solitaire games for Omarchy, offline and ad-free. Omarchy
  10. ZenPDFA clean, open-source PDF workbench with practical tools and transparent usage limits. Web App
  11. ZenPDF for OmarchyAn offline-first Acrobat alternative for Omarchy and Arch Linux. In Progress

OSS Contributions

A selection from popular projects at Google, Meta, Microsoft, Grafana Labs, T3 and others. These aren't all of them: see more on GitHub.

  1. google/go-containerregistryGoogle4.1k ★
    • #2316Kept localhost/... image references pointing at the local registry.
    • #2315Let push-only workflows skip Kubernetes pull secrets.
    • #2314Ensured only true loopback hosts get plain-HTTP registry access.
  2. meta/lexicalMeta24k ★
    • #8539Fixed collapsible blocks breaking on empty or inline-only content.
    • #8541Added markdown round-trip tests for bold-italic text.
  3. microsoft/onnxruntimeMicrosoft22k ★
    • #32846Enabled int8 and uint32 support in the CPU Where operator.
  4. grafana/k6Grafana Labs32k ★
    • #6071Added static labels so Prometheus metrics from parallel k6 runs stay distinguishable.
    • #6070Stopped recent console logs getting lost by flushing file logs during runs.
    • #6002Added single() to grab the first matching element in k6/html.
  5. pingdotgg/t3codeT332k ★
    • #12768Made desktop builds report one consistent version across the client, server and package.
    • #19923Stopped large image attachments stalling the chat timeline by decoding them asynchronously.
  6. cel-expr/cel-goGoogle3.1k ★
    • #1340Added a JSON encoder to CEL's standard extensions.
    • #1327Allowed container aliases for simple names, with validation still strict.
  7. google/osv-scannerGoogle11k ★
    • #2808Removed duplicate vulnerability queries in bulk scans, keeping result order.
  8. meta-pytorch/botorchMeta3.6k ★
    • #3370Fixed sampling from batched truncated normal distributions, which used to reject batched constraints.
    • #3372Let nonlinear-constrained optimization start from infeasible points, since SLSQP can repair them.
  9. meta/docusaurusMeta67k ★
    • #12054Fixed localized sites publishing wrong URLs in metadata and sitemaps.
  10. meta-pytorch/torchftMeta552 ★
    • #356Made checkpoint transfers honour the transport's configured timeout by default.
  11. microsoft/playwright-pythonMicrosoft15k ★
    • #3084Fixed handlers with keyword-only parameters getting the wrong arguments.
  12. Azure/azure-cliMicrosoft4.7k ★
    • #33429Kept az vm image list working when the remote alias file is malformed.
  13. openclaw/gogcliOpenClaw8.5k ★
    • #555Added a safe preview for finding duplicate Google Contacts.
    • #554Added read-only Drive reports: tree views, folder summaries and inspection.
  14. aaif-goose/gooseAgentic AI Foundation56k ★
    • #12516Kept session context alive across the agent's reply lifecycle, including ACP delegation and resumes.
  15. Unstructured-IO/unstructuredUnstructured16k ★
    • #4504Kept element IDs and source metadata intact when reloading Unstructured's own JSON output.

Work

In ML since 2019
  1. Jun 2025 – Now

    Lead ML Engineer Altimetrik

    Feature engineering and ML orchestration for Calibo's data and AI platform, which runs workflows across Databricks and Snowflake. I own technical design, delivery and integration across its ML services and data pipelines.

    Lead ML Engineer(prev. Senior ML Engineer)

    • Feature engineering track: built configurable data preparation in Python, PySpark and SQL (source filtering, missing-value handling, query profiling and custom features), with LangGraph review workflows behind FastAPI that validate each plan before approved configs move to the next stage.
    • ML orchestration and training: built reusable scikit-learn training for six classification and regression algorithms with held-out evaluation and saved inference pipelines, made every run traceable in Databricks and MLflow, and added pytest regression coverage for bad feature and preprocessing settings.

    Python, PySpark, SQL, Databricks, Unity Catalog, MLflow, scikit-learn, LangGraph, LangChain, FastAPI, Pydantic, pytest

  2. Apr 2024 – Oct 2024

    Data Scientist National Payments Corporation of India

    Led fraud detection for AePS, turning 4TB+ of raw transactions into features and predictive models that cut fraud incidents by 1.2 million. Built supervised time-series anomaly detection, and a federated fraud model for banks newly joining UPI.

  3. 2022 – 2023Co-founder, own venture

    Data Scientist SkipCheckr AI

    Took an ML-powered retail self-checkout from concept to MVP with computer vision, targeted research and A/B testing.

  4. Jul 2019 – Apr 2022

    Data Scientist & Software Engineer Siemens

    Fault detection on client assets from sound and sensor data, reaching 92% accuracy on remaining-useful-life classification, plus the dashboards that tracked the health of 97 assets.

Résumé as a PDF, full history on LinkedIn.

Writing

More on Medium

Contact

Open to: remote roles, or on-site in San Francisco, Hyderabad, Seattle, New York, Amsterdam, Copenhagen, Frankfurt and Munich.

The fastest way to reach me is a DM on X. I'm also on: