Metallurgical & Materials Engineering undergraduate at IIT Kharagpur with extensive hands-on experience in NLP, computer vision, and full-stack data applications. Proven track record of leading cross-functional teams to deliver scalable AI/ML pipelines, from research prototyping to production deployment. Passionate about building robust, maintainable systems that drive business impact.
Experience
Enhanced the NLP capabilities of 'Sara', a medical-AI NL-to-SQL converter for healthcare analytics.
- Fine-tuned GTR-T5 and Flan-T5 XL models on custom clinical datasets
- Built a RAG pipeline using LangChain, Milvus, and multi-vector embeddings
- Improved query accuracy from 72% to 87% and reduced latency by 30%
- Integrated FastAPI endpoints to serve model predictions at scale
Instruct-tuned Meta-Llama-3-8B to handle contextual data query workflows for enterprise analytics.
- Implemented 16-head, 500M-parameter model with mixed-precision and quantization
- Reduced memory footprint by 35% and training time by 20%
- Deployed model container on AWS SageMaker for A/B testing
- Authored internal guidelines for best practices in LLM fine-tuning
Supported data-driven research projects in business analytics and financial modeling.
- Cleaned and analyzed large economic datasets using Pandas and SQL
- Visualized trend analyses in Jupyter notebooks for publication in school reports
Developed an AI-driven startup portal and NLP-based SWOT analysis tools powered by Groq API.
- Engineered NLP pipelines for automated SWOT extraction, boosting decision efficiency by 40%
- Built RESTful FastAPI services with CORS and authentication
- Designed a React frontend prototype for user interactions
Projects
Real-time retrieval and summarization of Medium articles via a multi-model LangChain architecture.
- Integrated Anthropic Claude 3 and Hugging Face Flan-T5 XL for summaries
- Reduced average query latency by 40%, delivering results in under 5 seconds
- Built and deployed interactive Streamlit UI on Render with caching and authentication
Regression model predicting track popularity with R² of 0.85 and MSE of 3.2 on acoustic metadata.
- Performed PCA to reduce feature space by 35% while retaining 90% variance
- Trained Random Forest & Gradient Boosting models with hyperparameter tuning
- Identified energy & danceability as top drivers, boosting prediction accuracy by 25%
Personalized business strategy portal leveraging Groq API for high-throughput NLP inference.
- Engineered NLP techniques for SWOT extraction and SWOT-based recommendations
- Boosted decision-making efficiency by 40% through optimized prompt engineering
- Built scalable FastAPI endpoints and deployed on AWS with CI/CD
Education
- Probability and Statistics
- Linear Algebra
- Machine Learning Foundations
- Programming and Data Structures