Senior Software Engineer - AI/ML

Machani Robotics | July 2024 - Present

• • Multilingual Speech Recognition: Developed a real-time speech-to-text pipeline using GStreamer and Whisper CPP, supporting 5 languages (English, Spanish, Italian, German, Portuguese) for diverse user interactions and seamless multilingual communication.
  
  
  • Whisper Fine-Tuning for STT: Fine-tuned the Whisper Base multilingual model using 40+ hours of Common Voice and LibriSpeech datasets, achieving a 15% reduction in Word Error Rate (WER) across all supported languages.
  
  
  • LLM Fine-Tuning with Phi3: Fine-tuned the Phi3 (3B) language model with 2000+ domain-specific Q&A pairs generated using GPT-4, improving contextual understanding and reducing hallucination in chatbot responses.
  
  
  • Retrieval-Augmented Generation (RAG): Designed and implemented a RAG framework combining LLMs with live face and voice input, enabling dynamic, context-aware conversations and enhancing real-time personalization.
  
  

SDE-I - AI Engineer

Machani Robotics | February 2021 - June 2024

• • Face Recognition Pipeline on Edge: Built a full face recognition pipeline using ArcFace (fine-tuned for Indian faces, 92% accuracy), DeepStream, and ONNX, optimized for low-latency deployment on NVIDIA AGX Jetson.
  
  
  • Real-Time Vector Search with MILVUS: Created an embedding storage and search system using FAISS and MILVUS, supporting fast and scalable vector-based lookup for face and speaker identity verification.
  
  
  • Custom Text-to-Speech System: Integrated OpenAI TTS and Cereproc APIs for voice synthesis and led custom TTS model development to support emotion and voice cloning, improving voice clarity and context-awareness by 20%.
  
  
  • Gesture Generation Using LLMs: Developed a deep learning-based gesture generation module using LLMs, improving the realism of body language and facial animation by 30%.
  
  
  • Optimized Edge AI Deployment: Engineered lightweight, ONNX-converted AI models with GStreamer pipelines, reducing inference latency for face and voice tasks by 20% on NVIDIA Jetson hardware.

Deep Learning and AI Instructor (Part Time)

AnalytiixLab | April 2022 - Present

• • Successfully completed 8 batches and trained over 150 students in Deep Learning and AI basics.
  
  
  • Also, conducted corporate training with Tredence Inc. and Samsung for the same.
  
  

Associate Data Scientist

Cognizant Technology Solution | April 2015 - January 2021

• • Developed and implemented advanced analytics solutions to drive customer insights and business strategies across various projects.
  
  
  • Developed predictive models to identify potential churn customers, enabling a 15% improvement in retention and supporting the creation of targeted promotional strategies.
  
  
  • Prioritized high-revenue leads using customer acquisition analytics, optimizing marketing resources and boosting acquisition efficiency by 20%.
  
  
  • Conducted sentiment analysis on 10,000+ consumer reviews, delivering actionable insights that directly influenced product and marketing strategies.
  
  
  • Implemented machine learning models to classify driver behavior from 2D dashcam images, improving safety outcomes and reducing incident detection time by 25%.
  
  
  • Collaborated with cross-functional teams to deliver scalable, data-driven solutions and effectively presented analytical findings to key stakeholders.

B.Tech In ECE

West Bengal University of technology | August 2010 - July 2014

Higher Secondary

West Bengal Board of Higher Secondary Education | August 2007 - July 2009

Secondary

West Bengal Board of Secondary Education | July 2007

Certifications

• Wiley Data Scientist. Credential ID: CZN-CDS-BAN-210819004
• Extensive Vision AI program. from The School of AI
• Extensive Python AI program. from The School of AI
• Extensive NLP program. from The School of AI