Shreeyash Pawar
ML @ Philips Research | 2 Patents | Ex Société Générale | COEP | 5x AI hackathons wins
Bengaluru
9+ years of work experience
Oct 20242 yrsPresent
Crafting GenAI solutions from research to production.
- Integrated VLM's (Qwen-omni) for video ingestion in patient monitoring chatbot
- Computer use agent (CUA) for zero shot navigation of medical screens and courses
- Scaling test time inference in agentic text2sql tasks
Dec 20212 yrs 11 mosOct 2024
• Led a 4 people team that built an full-stack end-to-end AI vision annotation platform, cutting clinical imaging annotation time by more than 4×
Stack → Pytorch | AWS | EKS | FastAPI | SQLAlchemy | Docker | IaC/CloudFormation | Airflow
• Trained a model that reconciles many human annotations into one ground truth for medical segmentation, using statistical methods & deep learning (Probabilistic UNet)
Stack -> Pytorch | OpenCV | 3D Slicer | ITKSNap | Numpy | FastAPI
• Brought deep-learning to the browser for real-time, server-less insight at the edge for a X-ray imaging platform.
Stack → C++ | ONNX | TensorFlow.js | SIMD-powered WASM
• Gave clinicians back hours each week through a near zero-hallucination clinical text summarizer, stress-tested against human experts.
Stack → LangChain | LlamaIndex | Azure | statsmodels | DSPy | RAG
• Developed and benchmarked knowledge graph for document search and categorization.
Stack -> AWS Neptune | Neo4j | NetworkX | Gremlin
Apr 20201 yr 9 mosDec 2021
Bengaluru, Karnataka, India
Applied deep learning in life sciences domain.
Project: Optimizing Drug Formulations Using AI
• Developed an application that determines optimal concentration values for drug formulations using SciPy's minimize solver.
• Enhanced ML prediction pipeline performance by 90% through code profiling and optimization using JobLib parallel processing, Numba JIT compilation, cuML, cuDF, and multiprocessing.
• Built end-to-end Python application with robust data pipelines using DVC for data ingestion, MLflow for model tracking, Dask and H2O AutoML for training, GitLab CI/CD pipelines for deployment, and Docker registry for model serving.
Project: De Novo Molecule Generation
• Implemented generative modeling using unsupervised/semi-supervised deep learning methods (GANs, VAEs) and encoder-decoder models for novel drug discovery targeting specific proteins.
• Work resulted in 1 patent filing.
Infra: Set up AI training infrastructure with Ubuntu Server, configured GitLab Enterprise Server, standardized deployment pipelines, and provided Linux environment troubleshooting.
Jul 20199 moApr 2020
Greater Bengaluru Area
Python programming for internal tools
Jun 20181 y 1 moJul 2019
Pune, Maharashtra, India
Worked on Computer Vision for Embedded Linux platforms.
• Developed real time scalable face recognition system that can run on constrained ARM devices.
• Successfully piloted “AI Face registration” at “Women in DS, conference’19”,a Stanford initiative, at Pune.
• Achieved record FPS on Arm v7 architecture using model optimisation techniques and scrutinized the effect of various tuning parameters on a standard dense neural network
Jan 20171 y 4 moMay 2018
India
The team builds a hybrid trike, mechanically operated by pedaling and electrically driven by a motor, powered by a battery. I lead electrical and electronics subsystem.
Shreeyash’s AI Stack
Education
Bachelor of Technology - BTech, Electrical, Electronics and Communications Engineering
COEP Technological University
20154 yrs2019
Join Shreeyash and 10k+ other professionals on
Tal is the app where bosses meet engineers & builders directly, without middlemen
Download Tal Boss