Experience

Computer Vision Intern

Vaelsys

Madrid, Spain

Supervisor: Dr. Eduardo Cermeño

• Built an edge video threat-detection pipeline for the Vaelsys–Intel proof of concept, benchmarking 5+ VLMs and ML classifiers with up to 95% theft recall.
• Benchmarked Qwen3-VL-8B inference on Intel iGPU through OpenVINO Model Server and gRPC, supporting 8–12 concurrent camera streams.
• Stress-tested an NPU–GPU two-stage edge-AI architecture combining D-FINE, DLStreamer, GStreamer, VLM inference, and ML classification.
• Led a VLM reliability study on counterfactual anomaly-region grounding for industrial inspection across 13k+ evaluations.

GenAI Engineer

Vionix Biosciences

USA · Remote

Supervisor: Dr. Jagadish Venkataraman

• Led RAG pipelines for spectroscopy Q&A, integrating embeddings, Milvus, and automated cloud indexing.
• Built asynchronous APIs and NL2SQL pipelines for automated metadata and file retrieval with secure cloud access.
• Recognized by executive leadership for high-impact delivery and offered a permanent role.

Data Scientist

SoftTechNation

Remote

• Developed a RAG web application with AstraDB and Groq using Llama embeddings for efficient loading, chunking, and storage.
• Worked across MLOps and ETL with Docker, PySpark, Airflow, and dbt.

Analyst

UrbanLyfe

Gujarat, India · Remote

• Built a Botpress chatbot for automated product queries and explored LLM fine-tuning for improved responses.
• Explored applied ML and BI techniques to integrate retail stores and optimize logistics sustainably.

Research Intern

University College Dublin & University of Cincinnati

Remote

Supervisor: Dr. Soumyabrata Dev

• Developed a fusion encoder–decoder model for emotion-enriched image captions with pretrained deep-learning features.
• Fine-tuned BERT and RoBERTa for adversarial prompt detection to prevent LLM jailbreaking.
• Performed sentiment analysis on TikTok reviews to infer psychological behaviour.

Project Trainee

Defence Research and Development Organisation

New Delhi, India

Supervisor: Sci. Gaurav Gupta

• Developed an offline speech-detection system with OpenAI Whisper, reaching 91% accuracy and 4.2% WER.
• Implemented speech-recognition to audio-segmentation conversion followed by WAV conversion for uninterrupted audio input.