Ashutosh Roy

AI/ML Engineer with hands-on experience building end-to-end LLM, GenAI, RAG, and speech AI systems. Published researcher (RECCAP 2026, IIT Palakkad) with a co-authored paper from IIT Delhi. Skilled in fine-tuning transformer models, designing hybrid retrieval pipelines, deploying production ML systems via FastAPI, and building intelligent automation workflows.

Research Intern - Indian Institute of Technology (IIT) Delhi - New Delhi, India

(2025-01 - 2025-06)

• Built an OCR-powered document understanding pipeline for archaeological datasets; benchmarked Tesseract, Marker, and Gemini OCR — selecting optimal strategy per document class.
• Designed hybrid retrieval (dense: FAISS/Pinecone + sparse: Apache Solr) for high-precision semantic + lexical search over thousands of historical records.
• Deployed scalable FastAPI services for real-time LLM-based querying with sub-second retrieval responses.
• Co-authored SARCH: Multimodal Search for Archaeological Archives — submitted for peer review. arxiv.org/abs/2511.05667

AI/ML Engineer Intern - MitoVoid AI - Remote

(2024-10 - 2024-11)

• Fine-tuned Mistral/LLaMA variants for healthcare NLP — medical record summarization and symptom-driven conversational AI.
• Reduced patient query resolution time by ˜40% by deploying an AI chatbot for telemedicine with optimized intent understanding.
• Built end-to-end ML pipelines: data preprocessing → model inference → API deployment for real-time healthcare applications.

B.Tech (Hons) - Computer Science and Engineering (Artificial Intelligence) - Chhattisgarh Swami Vivekananda Technical University (CSVTU) (2022 - 2026)