Production Service Engineer(Gen AI Engineer)
Zoom Communications
- Built Applied AI agentic chatbot with LLMs and Semantic Kernel to analyze customer issues, process transcripts and generate insights, reducing manual effort by 30%.
- Architected and deployed scalable retrieval-augmented generation (RAG)using LangChain and Databricks Vector Search across 22+ domains at sub-3s latency, improving retrieval accuracy and reliability.
- Deployed DistilBERT for real-time complaint classification with 95% accuracy and sub-1s latency.
- Built an XGBoost + Random Forest ensemble pipeline with a feature store and automated retraining, reducing contact rates by 35%.
- Generated 200+ actionable insights using LLMs, prompt engineering, few-shot learning and evaluation frameworks.
- Implemented multi-agent reasoning workflows with automated evaluation metrics and monitoring dashboards.
- Built an AI Hackathon prototype that generated 2-second virtual-background videos from live speech transcripts using LVDM as the base model and VideoLoRA for fine-tuning.
- Developed Zoom’s workforce management forecasting module with SARIMAX, Prophet and ensemble models; reduced RMSE by 23% versus the previous model and achieved 200x faster execution.
- Built Vista and Atlaz, AI tools for knowledge retrieval, ticket analysis and troubleshooting automation.
- Resolved 2,800 technical issues for 40+ enterprise customers; supported root-cause analysis and fix validation.
Machine Learning Intern
NVIDIA
- Fine-tuned Hugging Face Transformers for ASR punctuation and capitalization using Tatoeba and ParaCrawl; prepared 490 hours of Russian speech data and contributed to Conformer model training.
- Deployed WFST text normalization using Google’s Kestrel framework; achieved 0.1 word error rate (WER) and 0.22 character error rate (CER) with audio-based normalization.
VISTA
Built a conversational platform with 8+ specialized agents/tools, integrating Claude and Ollama through an LLM gateway with Next.js and MongoDB; secured access using Okta OAuth2/OIDC and RBAC. Unified tickets, bugs, release notes and troubleshooting knowledge; automated log ingestion, decryption and multi-file processing for self-service investigations, contributing to 22% fewer tickets over six months.
Atlaz | AI-Powered Ticket Intelligence
Retrieved Jira cases, knowledge articles and prior resolutions to recommend diagnostics and required logs. Extracted log errors and request types and surfaced diagnostics in Jira to accelerate root-cause analysis, improving issue resolution time by 28%.
Multi Organ Segmentation
Combined Transformers and U-Net (UNETR) to segment multiple organs in 3D medical images, achieving a Dice score of 0.759
B.Tech + M.Tech
Indian Institute of Information Technology, Kancheepuram · 9.12
Structuring Machine Learning Projects
DeepLearning.AI
Jul 2020Stanford University
Dec 2019Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization
DeepLearning.AI
Jul 2020Databricks
Oct 2026Coursera
Jun 2019
Attendee · Oct 24, 2026