Jun 2024 – Nov 2025 · 1 yr 5 mos · Bengaluru, Karnataka, India · Hybrid

• Accuracy Debugger Tool (Quantized Models)
• Contributed to the development of Accuracy Debugger, a tool for diagnosing accuracy degradation in quantized models
• Identified root causes of high quantization error in a quantized Google Nano model using the tool
• Optimized core debugging algorithms, significantly improving execution speed by enhancing model subgraph dissection
• Co-authored research poster: “Innovative Methods for Exploring AI Model Behaviour in Low Precisions” (QBUZZ 2024)
• QAIOR Tools & Model Fine-Tuning Frameworks
• Led development of the QAIOR tools repository for fine-tuning models across multiple tasks
• Implemented data parsing, extraction, and chunking for diverse document formats
• Built dataset generation pipelines using LLM prompting with structured output parsing
• Developed fine-tuning frameworks for RAG systems, integrating LoRA, QAT, and two-tower retrieval architectures
• Enabled fine-tuning for function-calling workflows, covering intent classification, parameter extraction, and response generation
• Designed end-to-end evaluation pipelines to validate model performance across tasks

Jun 2022 – Jun 2024 · 2 yrs · Delhi, India

• Machine Learning Engineer
• Professional Achievements:
• 1. Led the development of a user proximity detection system using Wi-Fi signals, resulting in the award-winning "Best Innovation Stellar Project Award" in 2023.
• Pioneered data collection encryption algorithm design.
• Innovated signal processing-based data preprocessing pipeline development.
• Developed the end-to-end process, including deep learning model training, testing, and inference pipeline development.
• Deployed an on-device solution in C++ using TensorFlow.
• 2. proposed the idea "Human Activity Detection using Wi-Fi for Home Monitoring" at the SRID Hackathon, where I served as the team leader. Our project was among 500 competing ideas and won the "People's Choice Award" at the SRI-Delhi Hackathon in 2022.
• 3. Optimized mmWave motion detection services by designing a cost-effective, proprietary algorithm that directly utilizes raw sensor data, obviating the need for specialized microcontrollers and vendor chips.
• Awards:
• Secured the "People's Choice Award" at the SRI-Delhi Hackathon in 2022.
• Recognized with the "Best Innovation Stellar Project Award" in 2023.
• Granted 3 patents (2 A1, A2).

Jun 2021 – Jul 2021 · 1 mo · India · Remote

• work
• 1. Development of Acoustic Echo Cancellation Solution based on deep learning.
• Implementation of DTLN-AEC training & augmentation pipeline (https://arxiv.org/pdf/2010.14337.pdf)
• Inference Pipeline Development for tflite module for deployment
• 2. Development of Sound Event Detection Solution based on deep learning based Student-Teacher Model
• Model capable of training on Strongly labelled and weakly labelled data using student teacher architecture