POOJA BHAVSAR

Available to hire

I’m a Generative AI/ML Engineer with 3+ years of experience building production-scale LLM, RAG, agentic AI, and Voice AI systems.

About

I’m a Generative AI/ML Engineer with 3+ years of experience building production-scale LLM, RAG, agentic AI, and Voice AI systems. In my current role at Sterlite Technologies (STL), I architected and shipped an agentic multi-agent IVR/call-routing platform that serves 1,000+ daily interactions per multi-tenant setup, delivering a 90%+ reduction in IVR latency while improving GPU memory efficiency by ~60%. I also built multilingual RAG pipelines to support 10+ low-resource Indian languages and reduce unanswered queries, using automated document ingestion and smart retrieval matching.

Alongside LLM systems, I’ve led the end-to-end delivery of a multilingual ASR–TTS web platform, including post-call transcript generation, speaker diarization, call summarization, and sentiment analysis for contact-center workflows. I enjoy turning complex ML and data-quality problems into reliable production solutions, mentoring a small team, and helping others by advising on diagnostics and model feedback loops—always aiming for systems that are accurate, efficient, and scalable.

  • Joined October 2023

Languages

  • EnglishFluent
  • HindiFluent
  • GujaratiFluent

Industry experience

  • Government
  • Computers & Electronics
  • Other
  • Software & Internet
  • Education
  • Professional Services
  • Telecommunications
  • Healthcare
  • Media & Entertainment

More experts

Projects · 2

Engineered a Plant Disease Detection CNN in PyTorch from scratch during a 15-day program, achieving 90% accuracy via Computer
Vision, deep learning architecture design, and transfer learning fine-tuning.
• Boosted accuracy by 10% over baseline through image preprocessing using OpenCV (histogram equalization, edge detection) and
Reinforcement Learning-inspired hyperparameter optimization.
• Resolved class imbalance using oversampling and focal loss, demonstrating proficiency in deep learning fine-tuning, image augmentation,
and production-ready model packaging

Plant Disease Detection
  • AI Engineer

• Built an end-to-end DGA detection system: extracted n-gram/entropy features from 1M+ domain samples, trained a character-embedding
LSTM in TensorFlow/Keras achieving 98.4% accuracy and 0.988 ROC-AUC on 200K+ evaluation samples.
• Built a real-time Kafka producer-consumer pipeline streaming Zeek DNS logs (50,000+ entries/session) into a FastAPI inference endpoint,
Dockerized for continuous scoring; improved cyber threat detection rate by 40% and cut manual log-review effort by 60%.

DGA domain detection
  • AI Engineer

Work experience· 8 roles

AI Developer — Generative AI & LLM Systems · Sterlite Technologies (STL)Dec 2023 – Jun 2026 · 2 yrs 7 mos

Architected and implemented a multi-tenant agentic multi-agent IVR system (Asterisk, MCP orchestration, RTP gateway, LangChain/LangGraph, STT/TTS) to eliminate manual call routing. Achieved 90%+ IVR latency reduction and reduced peak GPU memory by ~60% using an MoE-inspired routing approach with prompt-engineered intent routing on a 4-bit NF4 quantized Llama-3.2-3B backbone. Built a scalable multilingual RAG pipeline supporting 1,000+ daily tenant interactions with per-tenant vector stores and automated ingestion of PDF/DOCX/PPTX/CSV. Reduced unanswered queries across 10+ low-resource Indian languages using IndicTrans2 translation, MMR retrieval, and RapidFuzz-based FAQ matching, while reducing peak GPU memory by ~60%. Extended a multilingual ASR-TTS full-stack web platform with transcript generation, speaker diarization, call summarization, and sentiment analysis, using QLoRA fine-tuning and integration across the agentic pipeline.

AI Developer – Generative AI & LLM Systems · Sterlite Technologies (STL)Dec 2023 – Present · 3 yrs

Architected a production Agentic AI IVR system (Asterisk, MCP Server, RTP gateway, LangChain/LangGraph orchestration, Chainlit UI, STT/TTS), cutting call-handling latency from multi-second to sub-100ms and enabling multi-tenant scalability. Built a multi-tenant ConversationalRetrievalChain RAG pipeline (LangChain, ChromaDB, Ollama embeddings) with per-tenant vector stores and MMR retrieval (k=10, fetch_k=50), serving 1,000+ daily Q&A interactions with sentiment analysis on post-call transcripts. Deployed quantized Llama-3.2-3B-Instruct (4-bit NF4, 8-bit fallback) with async inference and CPU fallback, reducing peak memory by ~60%. Delivered bilingual English/Hindi AI assistant with IndicTrans2 for real-time translation and template matching via RapidFuzz.

AI Engineer (RAG-based Q&A System) · Sterlite TechnologiesDec 2023 – Present · 3 yrs

Built a RAG-based Q&A system enabling accurate information retrieval from structured and unstructured data. Developed IVR flows with Asterisk, speech-to-text, and AI-powered responses using LLMs; created a transcription module for dial centers and multi-tenant chatbots/voice bots; enhanced user interaction and summarization.

AI/ML Trainer · Grass Solutions Pvt LtdNov 2023 – Dec 2023 · 2 mos
AI/ML Intern — Cybersecurity · ISRO – Space Applications CentreJan 2023 – Jun 2023 · 6 mos
AI/ML Intern — Cybersecurity · ISRO – Space Applications Centre (SAC)Jan 2023 – Jun 2023 · 6 mos
Machine Learning Intern – Cybersecurity · ISRO – Space Applications CentreJan 2023 – Jun 2023 · 6 mos
Machine Learning Intern · Maxgen TechnologiesJun 2022 – Jun 2022 · 1 mo

Education· 6

bachelor of computer engineering · silver oak university2019 – 2023
Bachelor of Engineering (BE) in Computer Engineering · Oak College of Engineering and Technology2017 – 2019
Higher Secondary Certificate (12th) · SC Higher Education2030 – 2026
B.E. Computer Engineering · Silver Oak College of Engineering & Technology2019 – 2023

Certifications and awards

Prompt Design in Vertex AI2030 – 2026
Inspect Rich Documents with Gemini Multimodality & Multimodal RAG — Google Cloud Skills Boost2030 – 2026

Hire POOJA BHAVSAR today

Post a job and invite POOJA, or contact them directly. No hidden commission.

Post a job