Abhirup Adhikary

I’m a final-year CSE undergraduate at IIT (ISM), Dhanbad, focused on turning AI research into reliable, production-grade systems. I’ve consistently performed at the top level in data and realtime LLM environments, including Top 0.2% in a Kharagpur Data Science Hackathon and Top 5 in a Realtime LLMs Bootcamp.

In my AI Engineering Intern role at Miror Ventures, I rescued a failing Retrieval-Augmented Generation pipeline by diagnosing metadata inconsistencies and adding context-aware prompt injection and guardrails. Within a 1.5-week sprint, I improved retrieval coherence from <5% to >99%, and I also built an end-to-end IDP pipeline using OCR and LLMs that reduced manual review time by 40%.

At Samsung Research, I worked on computer vision and data pipelines—fine-tuning state-of-the-art architectures for juvenile gender detection and improving classification accuracy by 12–18%. I also developed a parallelized image preprocessing pipeline using Python and RetinaFace, processing 50,000+ images and increasing ingestion throughput by 3x.

Earlier at Capgemini, I designed and evaluated a Generative AI model to simulate optimization path discovery, then containerized and deployed it via a scalable FastAPI endpoint using Docker. I carry this same engineering mindset into my projects too—like a realtime stock news assistant (using Pathway and Dockerized OpenAI API endpoints) and time-series predictive models built with SARIMA and LSTM.