T. Pranav Email & Phone Number

Singapore

Singapore

AI for MTL Education | Engineering Human-AI Multimodal Interactionhttps://aisingapore.org/learn-language-automated-evaluation-by-generating-answers-questions-from-cartoons/

Singapore

Manager: Donny Soh Cheng Lock 1. Currently, leading the development of Visual Dialogue Engine comprising visual LLMs for Chinese, Malay, and Tamil, tailored for Singaporean primary school students.2. Used GPT-4 and cartoon images created by artists from AiSG and the Ministry of Education to curate datasets for general pre-training, fine-tuning, and instructional tuning of LLM series such as Gemma, LLaVa, and Qwen.3. Deployed several open-source LLMs and fine-tuned models to benchmark results on inference data.4. Engineered automated pipelines and API integration using Docker, NGINX, and Linux automation features, and conducted extensive evaluations involving 300+ students and parents to assess user-friendliness and interaction satisfaction.

Delhi, India

Manager: Hemant Yadav & Rajiv Ratn ShahWorked on fine-tuning a multi-speaker, multi-language TTS model for code-switching across 7 Indic languages. Achieved 6th place in the ICASSP LIMMITS 2024 track 1 Challenge for speaker similarity MOS.https://sites.google.com/view/limmits24/results

Gurugram, Haryana, India

Manager: Jaskaran and SharathWorked on speaker recommendation for multi speaker video dubbing, cross lingual voice cloning and OpenAI whisper algorithm engineering and validationhttps://dubverse.ai/subtitles/

Singapore, Singapore

End to End Robust Speaker Diarizationhttps://www.dropbox.com/scl/fi/ix2zecra23egczjk83x43/Interns-Jan-June2023.pdf?rlkey=v1fddv3g0n5etibyqsbskndtc

New Delhi, Delhi, India

Supervisor: Dr. Vinayak Abrol in collaboration with IDIAP Research Institute, SwitzerlandDevelopment of low footprint speaker verification models for smart devices

Noida, Uttar Pradesh, India

Manager: Tamoghno Bhattacharya & Parvesh JainBuilt an automated OCR solution for fin-tech from data labelling and annotation to detection of relevant data using Azure Form Recogniser. Deployed the model into a website using React.js and Node.js with the help of azure service. Also got an opportunity to explore few python libraries build for extracting texts from documents.

Remote

Supervisor: Dr. Sirisha RambhatlaImplemented baselines for time2vec and experimented it on stock price prediction on yahoo finance data. Also, proposed 2 research ideas, one on a universal time series representation, like BERT in NLP, and another on universal data encoder using cross-modal masking based generation after carrying out literature survey on multimodal learning, self supervised learning and time series representation in a group of two.

Patiala, Punjab, India

Supervisor: Dr. Tanu Wadhera & Dr. Neeraj KumarDeep learning in ophthalmology & Implemented vision transformers for classification and detection of breast cancer and skin lesions using MRI Datasets from kaggle and UCI ML repository.

Be- Electronics, Instrumentation And Control

Bachelor'S Thesis, End To End Robust Speaker Diarization, A

Part of speech and language lab, NTU under Prof. Chng Eng Siong and Dr. Lim Zhi Hao https://www.dropbox.com/scl/fi/ix2zecra23egczjk83x43/.

Cbse, Xii, 89.6 % | Grade: X, Gpa : 9.8/10