Headline

Data Scientist

About

As a Speech Data Scientist , I focus on building innovative data pipelines and optimizing machine learning models. My role involves applying advanced techniques in pipelines, to enhance the efficiency of speech systems.

Location

Bengaluru

AI stack

claude

gemini

hugging face

huggingface

Qwen

vLLM

Work experience

Speech Data Scientist at Contiinex, Mar 2026 - Present · 6 mos, Remote

AI/ML Engineer at Kiksy.ai, Sep 2025 - Mar 2026 · 7 mos, Bengaluru, Voice AI Platform: Designed and deployed a real-time Voice AI calling platform featuring turn detection, VAD-based interruption control, and bidirectional WebSocket streaming. Built Machine Learning Pipelines with different SOTA models in ASR,TTS,LLMS, and other Diffusion Based models. optimized the models at inference for faster inference Scalable Backend: Architected a horizontally scalable call scheduling system using FastAPI, Celery, and RabbitMQ, deployed via Docker to support multi-tenant campaigns. Real-time AI Avatars: Built an interactive AI Avatar system integrating LLM reasoning, RAG pipelines (using Qdrant and FAISS), and GPU-optimized lip-sync generation.

AI/ML Engineer at Tata Elxsi, Jun 2023 - Sep 2025 · 2 yrs 4 mos, Bengaluru, Mobile Deployment: Exported optimized models to TensorFlow Lite (TFLite) and integrated them into an Android application, achieving real-time inference with <200ms latency. Large-Scale Broadcast Analytics: Architected a 24/7 multi-channel video analysis platform that achieved ~90% automated tagging accuracy. Production Anomaly Detection: Developed and deployed an unsupervised defect detection system using Variational Autoencoders (VAE) for spark plug inspection, achieving ~87% accuracy on the production line. Edge AI & Model Optimization: Optimized a YOLOv5-based segmentation model for medical imaging, applying INT8 quantization and pruning to reduce model size by 4x with a minimal ~1% accuracy drop. Agentic Video Pipelines: Built an agentic generation pipeline using LangGraph, orchestrating diffusion models and LLM reasoning to ensure visual consistency across frames. Scalable Backend Engineering: Designed asynchronous backends using FastAPI, Celery, and RabbitMQ to process 1 hour of high-definition video in under 20 minutes. Industrial ML Innovation: Developed a sensorless payload mass estimation pipeline using PySpark and CAN bus telemetry, reducing hardware costs by leveraging predictive modeling.

Education

integrated masters of science, Physics, Sardar Vallabhbhai National Institute of Technology, Surat

Diploma of Education, Data Science, Indian Institute of Technology, Madras