Kaushal Paneri

Senior Data & Applied Scientist · Microsoft
PDF

Senior Data & Applied Scientist blending research and production ML — 9+ years of software engineering and 7+ years specializing in large-scale applied ML.

Skills

Research:Causal inference, reinforcement learning, foundation models, off-policy estimation, counterfactual inference, large-scale recommender systems, time-series forecasting, Bayesian methods.
Tools:Claude Code, Python, C#, PyTorch, scikit-learn, ONNX, Azure ML, MLOps, BOTorch, Git, GitHub Copilot, SQL.

Experience

Senior Data & Applied Scientist · Microsoft Advertising
Sep 2024 – Present
  • Own the forecasting models that drive automated-bidding decisions across the entire marketplace — a foundational product contribution powering advertiser spend decisions at scale.
  • Drive bid-recommendation and performance-estimation pipelines at the core of the automated-bidding stack — influencing ~$15M in annual advertiser spend.
  • Lead novel research combining foundation-model forecasts with model predictive control to make automated bidding more adaptive and efficient — targeting publication at a top ML venue.
  • Established CI/CD and security engineering for the team's ML codebase — automated testing and validation on every change; identified and resolved critical dependency vulnerabilities that could have impacted production.
Data & Applied Scientist 2 · Microsoft Advertising
Sep 2020 – Aug 2024
  • Published SigIS (Simulator-Guided Importance Sampling) as first-author oral at ACM RecSys CONSEQUENCES 2024 — a new method for tuning large-scale recommenders by combining simulation with importance sampling. Co-authored companion MixtureIS paper at the same venue.
  • Owned and modernized the production click-calibration ML stack for 4+ consecutive years, evolving it across generations (Probit → Random Forest → ONNX → full MLOps) and designing a generalized feature-extraction architecture that became the platform-wide default.
  • Launched and grew an auction-simulation product line from research prototype to production as primary owner — reached $1M+ annual advertiser spend influence in the first 7 months post-launch.
  • Brought Causal Transfer Random Forest (CTRF) from research paper to production click-prediction pipelines; co-presented the production integration at a Microsoft Research seminar with the original authors.
  • Mentored 3 summer research interns over 4 years — one project productionized into $400K annual savings, another converted into the peer-reviewed SigIS paper.
Data & Applied Scientist · Microsoft Advertising
Sep 2019 – Sep 2020
  • Delivered a stalled cross-team counterfactual-evaluation project within the first 3 months — ramping up a distributed big-data platform and a large C#/F# codebase from scratch.
  • Built the transfer-learning infrastructure (local comparison environments, distributed training, hyperparameter optimization) that directly enabled the team's subsequent migration from the legacy click-prediction model to Random Forest in production.
Visiting Lecturer · Northeastern University, Khoury College of Computer Sciences
2022 – 2024
  • Taught Supervised Machine Learning (graduate-level, 2 offerings) and Data Analytics and Machine Learning II (undergraduate).
Researcher · TCS Research Lab, Delhi
2014 – 2017
  • Designed and built a Bayesian Network Inference Engine adopted across 20+ client organizations.
  • Published 7 peer-reviewed papers (IEEE VIS, VAST, FUSION, ICDMW), won 4 international data challenges, and produced research output that contributed to 2 granted US patents.

Open Source

crux — Agentic Tool Manager for Claude Code · github.com/crux-cli/crux
2025 – Present
  • Authored and solely maintain an open-source Python CLI that lets developers orchestrate agentic workflows on top of Claude Code — with sandboxed tool execution, OS-keychain secrets management, and MCP (Model Context Protocol) server + skills integration.

Publications & Patents

  1. System and Method for Visual Bayesian Data Fusion · Patent US 10,430,417
    2019
  2. Multi-Sensor Visual Analytics · Patent US 10,013,634
    2018
  3. Regularizing Fully Convolutional Networks for Time Series Classification by Decorrelating Filters · AAAI 2019
    2018
  4. Speeding Up DNNs Using HPL Based Fine-Grained Tiling for Distributed Multi-GPU Training · Boston Area Computer Architecture Research Workshop 2018
    2017
  5. Visual Statistical Analysis of Environmental Sensor Data · IEEE VAST 2017
    2016
  6. Crop Planning Using Stochastic Visual Optimization · IEEE VDS 2017
    2016
  7. Visual Bayesian Fusion to Navigate a Data Lake · IEEE FUSION 2016
    2015
  8. Bayesian Visual Analysis of the Indian Labour Market · CoDS Data Challenge
    2015
  9. Multi-Sensor Visual Analytics Supported by Machine-Learning Models · IEEE ICDMW 2015
    2014
  10. Spatio-Temporal Analysis of Ethnic Conflicts and Human Rights Violations in Africa and the Middle East
  11. Leveraging Structured Biological Knowledge for Counterfactual Inference: A Case Study of Viral Pathogenesis · IEEE Transactions on Big Data (co-first author)
    2020

Professional Service & Peer Review

Education