Abhinav Goyal

I am a part of the Large Language Models team. As a part of this, I am building fast and efficient in-house LLMs for various e-commerce use cases. I am also working on Spoken Language Models which combine both text and speech modalities. Previously I led Flipkart's Speech Recognition to build ASR for various use cases, domains, and languages. As a part of this role, I built in-house ASR models for Indian languages which power Flipkart's Voice Search. In the past, I've also worked on augmenting the in-house translation models to infer word alignments, constituency parsing, and NLU tag transfer. I've also implemented solutions for entity classification and grapheme-to-phoneme conversion.

Developed Machine Reading system to answer comprehension based factual questions using Tesseract-OCR and pre-trained RNN models. Used Stanford-NLP for extracting entities and relations from comprehension. Trained CNN models in keras using supervised learning and transfer learning to recognize emotions from face images. Used opencv to detect faces and trained DNNs to recognize emotions from facial landmarks. Automated the detection and classification of defects in TV Video Stream by using statistical anomaly detection methods like moving average and rolling standard deviation.