- Worked on building robust conversational AI systems for Indian languages leveraging open-source LLMs, ASR, TTS and Language Modelling frameworks on EC2 machines and Azure VMs.- Deployed chat applications powered by open-source LLMs such as Llama-3 70b and Mixtral 8x22b using Langchain & LlamaIndex for conversational memory, code generation & interpretation. - Prepared, pre-processed/cleaned large-scale ASR and TTS Datasets of Indian Languages using innovative methods.- Trained, tested and deployed competitive ASR models using frameworks such as NVIDIA NeMo and K2-FSA Icefall as well as TTS models using StyleTTS2, MatchaTTS, Pflow, VITS and Piper, vocoders using VOCOS, HiFiGAN, etc.- Worked with relatively new architectures such as Conformer-CTC/RNNT, Zipformer-CTC/RNNT, Transformers, etc to achieve competitive performance.

- Conducted research studying and fine-tuning pre-trained CNNs for photovoltaic detection. - Compiled, prepared and pre-processed datasets for photovoltaic semantic segmentation and object detection. - Trained, tested and evaluated numerous CNN models on HPC clusters for geospatial machine learning research. - Presented "Investigating the viability of convolutional neural networks for photovoltaic detection using remote-sensing images" paper at InGARSS 2023.

Built ML models for sentiment analysis of customer reviews. Provided marketing leads through webscraping using Selenium. Trained and used automatic speech recognition models for call recordingsanalysis for marketing insights.