Open to opportunities

Anish Pawar

@anishpawar

Message

Machine learning scientist building continual-learning and LLM systems for real impact.

India

Message

What I'm looking for

I’m looking to build production ML systems where I can own end-to-end delivery—LLMs, computer vision, and forecasting—with measurable business impact, strong experimentation, and engineering support to ship reliably.

I’m an ML Scientist focused on turning advanced ML research into production-ready systems, especially for language understanding, continual learning, and document intelligence. I’ve worked on continual-learning token classification and built an MCP framework for zero-shot PHI pattern generation and evaluation.

In my current role, I adapted encoder models for token classification (92% accuracy), deployed a DINOv2 technical-figure detector (94% accuracy), and developed an LLM knowledge-distillation flow for document classification (90% accuracy). I also created LLM post-detection verification with optimized few-shot and chain-of-thought prompting (92% precision), delivering USD 200K+ in business impact.

Previously, as an Applied Scientist Intern, I benchmarked and fine-tuned Time-LLM variants and engineered features to reduce MAPE, built RAG for lag retrieval with graph and attention-based modeling (20% MAPE reduction), and designed an A2C-based geospatial clustering approach (30% MAPE reduction). I’ve also built practical implementations across computer vision and forecasting, including serverless ETL/inference that reduced stock-outs by 70%.

Experience

Work history, roles, and key accomplishments

Current

ML Scientist

Current

Matters.AI

Jul 2025 - Present (1 year)

Adapted encoder models for token classification in a continual learning setup and built an MCP framework for zero-shot PHI pattern generation and evaluation. Deployed a DINOv2 technical-figure detector and developed LLM knowledge-distillation and post-detection verification pipelines for document classification.

Token Classification Encoder MCP PHI Pattern Generation DinoV2 Technical Figure Detection Knowledge Distillation Document Classification Post Detection Verification

Applied Scientist Intern

Amazon

Jan 2025 - Jun 2025 (5 months)

Benchmarked and fine-tuned time-series forecasting models (Time-LLM, Autoformer, Lag-Llama, XGBoost, SARIMA), engineering features and transforms to improve MAPE. Built a RAG pipeline for lag retrieval and developed serverless ETL/inference for production data.

Project Intern

Electrical Engineering Department, IIT Bombay

Jan 2023 - May 2023 (4 months)

Benchmarked VSLAM algorithms for localization and mapping in agricultural environments. Integrated an Intel RealSense camera with a Jetson Nano on an RC vehicle to deploy real-time VSLAM mapping and navigation.

Localization Mapping Intel RealSense RC Vehicle Integration