Jul 2024 – Oct 2025 · 1 yr 3 mos

• Developed autonomous DeFi agents for portfolio management using risk optimization across blockchain protocols, scaling to 5M requests and managing $10M+ in assets with quantitative risk assessment frameworks.
• Engineered Langraph-powered risk-adjustment system with dynamic task routing, recalibrating investment strategies in real-time and achieving 25% improved portfolio performance.
• Fine-tuned Llama models and integrated API-based LLMs (OpenAI, Claude) for context-aware NLP in financial applications, enhancing decision-making with 200ms response times.
• Partnered with financial advisory firm to create a GPT-4 AI assistant for tailored investment advice, implementing constrained generation for recommendations.
• Built AI supply chain optimizer using AutoGen, GPT-4, and RAG for inventory management, reducing costs by 20% and improving delivery efficiency by 15%.
• Built comprehensive MLOps pipelines with Apache Airflow for automated model training and deployment, ensuring 95% test coverage and robust validation using AWS CDK.
• Implemented personal finance assistant using LangChain and fine-tuned Mistral AI for expense categorization and portfolio optimization.

Aug 2022 – Jul 2024 · 1 yr 11 mos · United Kingdom · Remote

• Developed and deployed evaluation pipeline using Llama3-8B with instruction tuning and preference alignment (ORPO), benchmarking and replacing models to improve accuracy by 15%.
• Built RAG agent with Custom Llama3-8B, FastAPI, LangChain, and Qdrant, achieving 30% improved query relevance in production.
• Engineered end-to-end ML pipelines with 4x model size reduction and 2x latency improvement via INT8 quantization, maintaining 95% accuracy in Docker/vLLM environments.
• Created a fashion sales forecasting system using CLIP embeddings and GPT-3 to analyze product data, predict trends, and boost quarterly sales by 18%.

Jun 2020 – Aug 2022 · 2 yrs 2 mos · Remote

• Trained,deployed and monitored NLP pipelines for entity extraction,
• classification(zero-shot and few-shot) and summarization using AWS sagemaker.
• Containerized(docker) machine learning model, pushed to ECR and deployed
• into production using lambda/k8s after code reviews in an agile manner
• Improved the accuracy in the range of 25% for computer vision models using DL model and 40% for NLP model using LLMs ,reduced the latency by 30% using quantization and pruning.
• Model and Data Versioning Automated Version Control and tracking of model
• versions, along with the data used to train it.

May 2019 – Jul 2019 · 2 mos · Himachal Pradesh, India

• Worked on landslide prediction problem in Mandi district.
• Publication Accepted - international conference on advanced computing 2022