Headline

FDE at Simplismart

Location

Kolkata

Work experience

FDE at Simplismart, Jan 2026 - Present · 9 mos, On-site

AI Engineer II at Vahan.ai, Sep 2025 - Jan 2026 · 5 mos, Bengaluru, Worked on the implementation of a chatbot agent utilizing Langchain & LangGraph with human-like persona capable of delivery rider onboarding and user nudging. Designed robust Prompt and Context Engineering frameworks to eliminate hallucinations in GPT-4 based personas, ensuring high-fidelity interactions. Developed end-to-end ML pipeline from EDA, feature engineering to deployment, creating a CatBoost prediction model to forecast rider onboarding likelihood based on various journey touchpoints. Developed a scalable voicebot using Pipecat, RabbitMQ, Redis and Milvus as vector database for job intent filtration of delivery workers

AI Research Engineer at Dubverse, Nov 2024 - Oct 2025 · 1 yr, Gurugram, Haryana, India, Developed faster inference endpoints, reducing latency from 10s to 1.2s while maintaining response quality. Developed and fine-tune Audio LLMs as part of a 5-person team, increasing model accuracy by 25% for industry-specific tasks. Built custom data preprocessing pipelines that enhanced training efficiency by 40% and reduced model convergence time. Spearheaded the development and deployment of a real-time, multilingual Audio diffusion system, supporting 50+ languages.

Research Intern at Global DA Solutions Private Ltd, Dec 2021 - Dec 2021 · 1 mo, Remote, Worked on End‑to‑end Spoken Conversational Question Answering system using Langchain and Pinecone vector database. Utilizing various finetuning strategies (LoRA, QLoRA) for deploying LLMs on mobile devices

Senior Research Fellow, Speech & NLP lab at Department of Electrical Electronics and Communication Engineering, IIT Dharwad, Sep 2021 - Oct 2024 · 3 yrs 2 mos, On-site, Researched and built proof-of-concept AI models in Speech Recognition and Text classification. Worked on developing Multimodal LLMs (3B+) integrating vision, speech, and text capabilities resulting in a 40% improvement in cross-domain task performance. Worked on model optimization techniques like Onnx and TensorRT to improve inference performance during deployment. Developed a novel distillation technique for multimodal-LLMs reducing model complexity by 95% with no impact to performance. Led a team of 4 engineers in developing a novel few-shot framework for LLMs, enabling rapid adaptation to new domains with 70% less training data

Education

Master's degree, Operations Research, National Institute of Technology Durgapur

Master, Operations Research, National Institute of Technology, Durgapur