Somya Jangir

I am a deep learning and AI researcher with hands-on experience building end-to-end pipelines for computer vision, speech, and document‑centric GenAI applications.

At Axory AI I developed a deepfake image detection pipeline, trained CNN models with transfer learning, and achieved 95% validation accuracy through systematic experimentation and robust evaluation workflows.

I have collaborated on speech research—curating multi-domain datasets, benchmarking ASR systems like Whisper and Wav2Vec 2.0, and performing detailed error analyses to uncover model failure modes and resource trade-offs.

My projects include a RAG-based AI agent for research paper learning with fine-tuned LLMs, vector DBs, and an interactive Streamlit interface, plus research on startup funding and medical image classification—demonstrating both applied engineering and empirical analysis strengths.