Shreya Boyane

Machine Learning Engineer with 2+ years of production DevOps experience and an M.S. in Artificial Intelligence from WPI (May 2026, GPA 3.9). Algoverse Research Fellow and co-first author of a mechanistic interpretability paper on latent chain-of-thought reasoning accepted as a poster at the ICML 2026 FoGen Workshop. Recognized in the Google Cloud x NVIDIA developer community with a Golden Ticket Contest honorable mention at NVIDIA GTC 2026 for inference-optimization engagement across three completed learning pathways. Data Science GQP MVP for a production multi-agent document-intelligence system built for an industry sponsor.

1. AI Trainer Fellow, Software Engineering and Multimodal AI

   Apr 2026 to present

   Handshake AI, San Francisco, CA (remote). Contract fellowship.

   Selected as a domain-expert fellow improving frontier LLMs and vision-language models for leading AI labs.

   • Evaluated 40+ structured tasks, each a side-by-side assessment of two frontier models (80+ LLM and VLM outputs graded for correctness, code quality, technical depth and reasoning soundness), delivering rubric-based expert feedback.
   • Authored challenging visual and compositional reasoning benchmark items with verified ground-truth answers over real-world charts, diagrams, infographics and scientific figures for chart and diagram VQA.
   • Developed domain-specific prompts for frontier LLMs across software engineering, machine learning and agentic reasoning, and conducted independent research to inform prompt design and rubric construction.

2. Machine Learning Engineer

   Aug 2025 to Dec 2025

   Crossing x WPI, industry-sponsored Graduate Qualifying Project, Worcester, MA (hybrid). Industry-sponsored project.

   Data Science GQP Most Valuable Player (MVP)

   Owned client-facing delivery of a production multi-agent document-intelligence system for H-1B immigration workflows.

   • Ran requirements discovery with the sponsor's engineers, translated ambiguous legal-workflow needs into acceptance tests, led weekly stakeholder demos, and owned end-to-end integration and handoff.
   • Built a four-agent LLM pipeline on AWS (Amazon Bedrock, S3 ingest, Lambda, Step Functions, event-driven triggers) that extracts structured fields into schema-validated canonical JSON and generates unified QC reports for Form I-129 evidence.
   • Normalized 45+ critical fields and executed 50+ deterministic rules per case with severity-tagged issues and traceable field-level decisions.
   • Cut petition processing from 3 to 4 days to about 5 to 10 minutes at 95 to 99% structural accuracy and 99 to 100% missing-document detection, with 85 to 95% of repeatable checks automated, manual verification reduced 70 to 90% and cost reduced 50 to 70%, freeing about 2,000 hours per mid-size firm per season.
   • Added a RAG fallback with RAGAS evaluation loops (faithfulness, context recall) and regression gating, plus a FastAPI evaluation service with Prometheus latency and error-budget metrics.

   Advisors: Prof. Bahman Moraffah, Prof. Chun-Kit Ngan, Prof. Fatemeh Emdad.

   Presented at the WPI 2025-26 GQP GRIE Showcase, 24 April 2026.

3. Graduate Assistant, Hypersonic Vehicles Project

   Jun 2025 to May 2026

   Worcester Polytechnic Institute, Worcester, MA (on-site). Part-time.

   Part-time graduate assistant on WPI's hypersonic vehicles research project.

4. Graduate Student Researcher, Generative AI and LLMs for Healthcare

   Jan 2025 to Aug 2025

   Worcester Polytechnic Institute, Worcester, MA (hybrid). Credit-based directed research mentored by Prof. Roee Shraga.

   Built a production-minded GenAI plus human-in-the-loop triage assistant with difficulty-aware reasoning, confidence calibration and auditable workflows.

   • Built confidence-aware triage orchestration in Python, PyTorch and scikit-learn with human-in-the-loop review, raising triage precision 20%, ESI assignment accuracy 25%, and cutting false escalations 40%.
   • Engineered a dual-channel decision flow separating high-risk escalation from routine guidance, reducing LLM hallucinations and miscommunication 50%, unnecessary handoffs 30%, and improving patient-provider alignment 2.3x.
   • Implemented difficulty-weighted training and GPT-style calibration scoring: errors down 30%, reliability up 35%, early critical-case detection up 60%, non-critical misclassification down 70%.
   • Added LLM guardrails, escalation rules, slice tests, calibration curves, drift checks and error-taxonomy trending; enforced reproducible runs with fixed seeds and deterministic loaders; partnered with clinicians to turn findings into decision guidelines and SOPs.

5. Graduate Assistant, Laboratory of Spaceflight and Planetary Exploration (LSPE)

   Apr 2025 to Aug 2025

   Worcester Polytechnic Institute, Worcester, MA (hybrid). Part-time.

   Developed an interactive data-driven web application and Python backend for spaceflight trajectory visualization.

   • Built the trajectory visualization app and its Python backend; presented a project poster.
   • Collaborated with lab members on execution and delivery and iterated on data visualization techniques.

6. Lead DevOps Engineer

   Jun 2022 to Jul 2024

   Persistent Systems, Pune, India (hybrid). Full-time.

   Top Talent 2024, Bravo Individual Award, Team Excellence Award FY2023, Bravo Team Award

   Owned CI/CD and cloud infrastructure across 20+ production projects.

   • Accelerated software delivery 95% by building CI/CD pipelines in Azure DevOps, Docker and Terraform with integrated automated testing and gated approvals across 20+ projects.
   • Standardized VM provisioning and cluster setup with Ansible and Terraform infrastructure as code, cutting manual configuration 70% and provisioning time 40%, including AWS S3, Lambda and DynamoDB codified in Terraform.
   • Improved system uptime 25% by deploying Prometheus and Grafana observability dashboards monitoring latency, queue time and build health in real time; cut cloud operating costs 30%.
   • Integrated ML-Ops into existing workflows, streamlining model deployment 50% through automation; shortened integration cycles 70% by coordinating release tooling across Angular and React frontend teams; reduced critical vulnerabilities with Snyk scans in CI/CD gates.
   • Restructured release pipelines and variable groups to manage more than 10 code repositories cleanly.

7. Software Engineer Intern (.NET)

   Jan 2022 to Jul 2022

   Persistent Systems, Pune, India (on-site). Internship.

   Best Intern of the 2022 batch among 20,000+ new joiners across India

   Built ASP.NET Web APIs and integrated Azure Machine Learning for predictive analytics.

   • Engineered scalable ASP.NET Web APIs with .NET Core and ADO.NET and optimized Microsoft SQL Server and MySQL queries for 30% faster load times.
   • Integrated Azure Machine Learning for predictive analytics; completed the GEMS (Graduate Engineers with Multidimensional Skills) training program.

8. Web Developer Intern

   Mar 2021 to May 2021

   Flavalon IT Solution LLP, Mumbai, India (remote). Internship.

   Built and launched brand websites with chatbots and learned SEO fundamentals.

   • Developed and launched five interactive websites per quarter on WordPress with CSS and JavaScript.
   • Integrated a chatbot on each site, increasing real-time engagement 40%; optimized MySQL structures and queries for faster load times; added ML-based personalized recommendations; performed website analysis and SEO optimization.

1. Latent chain-of-thought trajectories as dynamical systems

   Feb 2026 to May 2026

   AI Research Fellow, lead pipeline engineer, Algoverse AI Research Fellowship. Advisor: Jerome Francis.

   Selected for the competitive 12-week Algoverse fellowship; modeled latent chain-of-thought reasoning in LLMs as discrete-time dynamical systems and reproduced two latent-reasoning training methods end to end on GPT-2.

   • Applied Dynamic Mode Decomposition, PHATE, UMAP, t-SNE and PCA to 768-dimensional latent trajectories across 8,792 GSM8K and 500 MATH samples, producing 22 quantitative plots per run.
   • Reproduced CODI self-distillation (parallel teacher-student with a 3-loss objective) and the COCONUT 4-stage latent curriculum as LoRA adapters (rank 128, alpha 32) on stock GPT-2 with PyTorch, Hugging Face Transformers and PEFT; recovered 99.84% of test samples with a positional overflow guard.
   • Engineered the reliability layer: a multi-stage Bash and Python orchestrator with CUDA health polling that stabilized 14+ hour training runs, HDF5 resume-from-cache saving about 90 minutes per recovery, a GPU-resident Kraskov KSG-1 mutual-information estimator, and a safetensors-level checkpoint repair that preserved trained embeddings byte for byte.
   • Built a verification harness that passes 28 of 28 reproducible numeric claims on the canonical run.

2. FairPrep: demographic fairness for table union search

   Jan 2025 to May 2026

   Research Assistant, Worcester Polytechnic Institute. Advisor: Dr. Roee Shraga. BSF Grant 2024022.

   First adaptation of the FairEM framework to annotator-level fairness in table union search, with a complete statistical methodology stack.

   • Built 9 source modules across 5 experimental phases on a 480-row dataset (60 annotators x 8 questions x 4 survey versions).
   • Designed the statistical stack: 10 FairEM confusion-matrix metrics, permutation testing (10,000 shuffles), bootstrap confidence intervals (10,000 stratified resamples), Fisher exact tests, Benjamini-Hochberg FDR correction, and dual-track logistic regression (scikit-learn for interpretation, statsmodels for inference).
   • Surfaced statistically significant bias against non-native English annotators on hard questions (25% versus 44% accuracy, p = 0.031), robust across 5 of 5 cleaning strategies.

3. Where does physics live in vision encoders?

   Mar 2026 to May 2026

   Lead researcher, Independent research.

   Spatial probing of physical reasoning across the vision encoder, projection and LLM stages of three vision-language models, plus a factorial LoRA ablation to find the leverage point for physical intuition.

   • Trained linear and MLP probes on per-patch activations at four pipeline stages of Qwen2.5-VL-7B, InternVL 2.5-8B and LLaVA-OneVision-7B using Physion++ ground-truth physical properties.
   • Produced physics saliency maps and measured information loss across the projection bottleneck.
   • Ran a five-condition factorial LoRA ablation (encoder, projection, LLM, encoder plus projection, full model) evaluated on PhysBench, GRASP Level 2 and ConservationBench.

1. Worcester Polytechnic Institute

   Aug 2024 to May 2026

   Master of Science, Artificial Intelligence. Worcester, MA. GPA 3.9/4.0.

   • Data Science GQP Most Valuable Player (MVP), December 2025
   • Golden Ticket Contest Honorable Mention, Google Cloud x NVIDIA developer community, NVIDIA GTC 2026

2. PES Modern College of Engineering (Savitribai Phule Pune University)

   Jun 2017 to May 2022

   Bachelor of Engineering, Information Technology, Honours in Artificial Intelligence and Machine Learning. Pune, India. GPA 8.9/10.

   • Honours in Artificial Intelligence and Machine Learning
   • Final-year project sponsorship from TECHVED, October 2021

1. Interpreting Latent CoT Reasoning as Dynamical Systems

   Sabari Iyyappan Duraipandian*, Shreya Sanjay Boyane*, Manju Nagesh*, Jerome Francis, Archana Vaidheeswaran, Kevin Zhu

   Poster, ICML 2026 FoGen Workshop, May 2026.

2. Toward Fairness-Aware Human Feedback for Table Unionability

   Nina Klimenkova*, Shreya Sanjay Boyane*, Roee Shraga

   Submitted, DASHSys 2026, May 2026.

* Equal contribution.

Machine learning and AI

PyTorch, PyTorch Lightning, TensorFlow, Keras, scikit-learn, Hugging Face Transformers, PEFT and LoRA fine-tuning, LLMs, vision-language models, LangChain, LangGraph, multi-agent systems, RAG, prompt engineering, model evaluation and benchmarking, LLM-as-judge and rubric grading, diffusion models, CycleGAN, VAEs and GANs, NeRF, reinforcement learning, recommender systems, statsmodels

Computer vision and 3D

OpenCV, object detection and segmentation (Mask R-CNN, Detic, SSD), monocular depth estimation, pose estimation (OpenPifPaf, MediaPipe), structure from motion, visual-inertial odometry, sensor fusion, optical flow, visual question answering, chart and document understanding, Blender Python API, Unity

Inference and GPU optimization

CUDA, TensorRT, ONNX, TorchScript, INT8 quantization, mixed precision, CUDA graphs, kernel fusion, Nsight Systems and Nsight Compute, NVIDIA NIM

MLOps, cloud and DevOps

AWS (Bedrock, S3, Lambda, Step Functions, EventBridge, CloudWatch, Textract, DynamoDB), Azure DevOps, Azure ML, GCP, Docker, Kubernetes, Terraform, Ansible, CI/CD (Azure Pipelines, GitHub Actions), MLflow, Prometheus, Grafana, Snyk, infrastructure as code, Linux, Slurm HPC

Programming and data

Python, C++, C#, Java, SQL, R, MATLAB, Scala, Rust, Bash, PowerShell, TypeScript and JavaScript, React, Angular, FastAPI, Flask, PostgreSQL, MySQL, MongoDB, HDF5, JSON Schema

1. May 2026

   Poster accepted at the ICML 2026 FoGen Workshop, ICML 2026 Workshop on Foundations of Deep Generative Models

2. Mar 2026

   Golden Ticket Contest honorable mention, NVIDIA GTC 2026, Google Cloud x NVIDIA developer community

3. Dec 2025

   Data Science GQP Most Valuable Player (MVP), Worcester Polytechnic Institute

4. Feb 2026

   AI Research Fellow, Algoverse AI Research Fellowship

5. Apr 2026

   AI Trainer Fellow, Handshake AI

6. May 2024

   Top Talent 2024, Persistent Systems

7. Dec 2023

   Bravo Individual Award, Technology and Process Excellence, Persistent Systems

8. Jan 2023

   Team Excellence Award FY2023, Persistent Systems annual awards

9. Dec 2022

   Bravo Team Award, Teamwork and Dedication, Persistent Systems

10. Jun 2022

    Graduate Engineers with Multidimensional Skills (GEMS), Persistent Systems