Apr 2024 – Mar 2025 · 11 mos · Bengaluru, Karnataka, India · On-site

• 1. LLM Training

Jul 2023 – Mar 2024 · 8 mos · Bengaluru, Karnataka, India · On-site

• 1. Foundational Large Language Model (LLM)
• a. Integral member of the core team that built Krutrim - India's 1st Foundational Language Model that supports upto 12 Indian Languages
• b. Achieved superior token to word ratio by training tokenizers for various Indian languages, surpassing existing systems' efficiency.
• c. Implemented advanced techniques for cleaning and filtering Indian language data, surpassing the effectiveness of existing methods.
• d. Established a robust pipeline for deduplication of Indian language data, optimizing processes by leveraging existing methods.
• 1. Live Streaming with Multilingual Translation and Lip Syncing:
• a. Spearheaded video lip syncing part of live streaming project with translation and lip syncing in five different Indian languages, enhancing accessibility and engagement.
• b. Incorporated Voice Activity Detection (VAD), Automatic Speech Recognition (ASR), Lip Sync, and Live Streaming using FFMPEG to create a live multilingual translation and lip-sync video, broadcasted seamlessly in real-time settings.

Jul 2022 – Jun 2023 · 11 mos · Bengaluru, Karnataka, India · On-site

• 1. Automated Detection of Battery Cell Defects (2 months):
• a. Spearheaded the development of an automated system for detecting physical defects in battery cells.
• b. Implemented classification as well as detection to enhance accuracy and efficiency.
• c. Collaborated with Ola Electric Factory as well as Automation Team to ensure seamless integration.
• 2. Traffic Signal Detection Using YOLOV5 (2 months):
• a. Achieved an impressive 85% accuracy in the detection of traffic signals using YOLOv5.
• b. Collaborated closely with annotators to ensure high-quality labeled data for continuous training with daily annotated data to increase accuracy.
• c. Established a workflow utilizing the DataLoop annotation tool to automatically retrieve newly annotated images and facilitate automated model training.
• 3. Automatic Speech Recognition (ASR) with OpenAI's Whisper Model (2 Months):
• a. Managed the inferencing of OpenAI's Whisper Model across multiple GPUs for conducting transcriptions of call center calls.
• b. Implemented optimizations to enhance speed and resource efficiency.
• 4. Language Model Fine-Tuning for FAQ Responses (1 months):
• a. Fine-tuned open-source Language Models (LLMs) to provide accurate responses to frequently asked questions (FAQs).
• b. Leveraged LoRA (Low Rank Approximation) technique to effectively fine-tune models, enhancing model performance and efficiency.
• b. Leveraged PyTorch's distributed package (torchrun) for parallel training across multiple nodes.
• 5. Language Model Fine-Tuning for Chatbot in Customer Care (3 months):
• a. Successfully finetuned open-source LLMs to serve as a chatbot in a customer care environment for specific problems.
• b. Applied DeepSpeed's ZeRO (Zero Redundancy Optimizer) Technique(s) to enhance model training efficiency.
• c. Structured multi-turn customer care data to ensure the model behaves like a ChatBot rather than a simple QA bot.

Jul 2019 – Jul 2022 · 3 yrs · Hyderabad, Telangana, India

Aug 2017 – Aug 2018 · 1 yr · Bengaluru, Karnataka, India

Apr 2017 – Jun 2017 · 2 mos · Bengaluru Area, India