Reetam Biswas

Designed document parsing workflows with LLMs (Hugging Face Transformers, Docling LangChain) to extract procedural knowledge from unstructured text, achieving 85%+ extraction accuracy and reducing manual documentation review by 40%. Performed knowledge labeling with Roboflow , creating 1,000+ annotated samples for LLM fine-tuning , improving supervised model training accuracy by 12%.

Transformed complex datasets into insights using Python (Pandas, NumPy), SQL, Tableau/Power BI, enabling 22% faster decision-making and saving 10+ hours per week in manual reporting. Optimized LLM pipelines with PyTorch/TensorFlow, LoRA, and prompt engineering, boosting model performance by 18% and cutting inference time by 30%. Collaborated via Jira, Confluence, Git/GitHub to align analytics with business and client goals, resulting in a 30% reduction in project turnaround time.

Improved ML pipelin efficiency by 64% by automating end-to-end training, deployment, and inference using Docker, AWS S3/ECS . Designed and deployed a LayoutLM-based Generic Parser with Hugging Face Transformers to extract entities and key-value pairs from claim documents into a production-grade database that improved data retrieval speed by 35% . Achieved 90% model prediction accuracy, and accelerated client delivery cycles as Team Lead. Deployed an Email Intent Classification.

Recognized with “Highflyer of the Quarter – Q3 2021” for outstanding performance and impact during internship. Developed a classification-based Deduction Validity Predictor using Python, scikit-learn, Pandas, improving validation efficiency by 35% and saving 100+ analyst hours per quarter Built a regression-based Payment Date Forecasting model (XGBoost, NumPy, Matplotlib), achieving 92% accuracy, improving client planning and enhancing cash flow visibility by 20%.

Semester 1: Generative AI, Neural Network Theory and Implementation, Elements of Machine Learning, Database Systems Semester 2 (In Progress): Statistical NLP, High-Level Computer Vision, Causethical ML, Machine Learning