Associate MLOps Engineer at SERENT, Bengaluru, Karnataka

Consolidated CI/CD pipelines for 11 API repositories into a centralized repository to prevent unauthorized changes and ensure consistent deployments. Standardized Kubernetes deployments by migrating repetitive configurations into reusable Helm charts with separate dev/prod YAMLs. Containerized and deployed ML models on NVIDIA Triton within the Kubernetes environment. Built monitoring and alerting using Grafana dashboards and automated Google Chat alerts. Improved rules engine inference performance by deploying Redis as a caching layer, achieving a 98.7% improvement over the previous setup.

ML Intern at USEReady, Bengaluru, Karnataka

Built an automated HR screening system on Elementum (Snowflake) using RAG for resume–JD matching, scoring, and automated candidate email responses. Implemented a multi-agent contract RAG system with a supervisor–worker architecture (Search, Extraction, Comparison) and orchestrated intent routing with persistent conversation state. Designed a high-precision retrieval pipeline using metadata pre-filtering (NLP-to-SQL) and hybrid retrieval (BM25 + vector + RRF) to reduce hallucinations. Implemented OCR-enabled, section-aware document ingestion with recursive chunking and sampling for accurate extraction. Deployed on AWS EC2 with Langfuse observability and PostgreSQL-backed persistent storage.

ML-Ops Intern at Deepnet Analytics, Visakhapatnam, Andhra Pradesh

Annotated 100+ PPE images using CVAT to improve dataset diversity and model generalization. Developed custom preprocessing (aspect ratio normalization, cropping torso/top/bottom) to improve PPE detection accuracy by ~12%. Deployed CVAT and Task-Creator models on AWS EC2, integrating with RDS for database connectivity and using S3 for scalable storage. Built a Dockerized model versioning pipeline using Terraform to ensure reproducible deployments and reduce setup time by 40%.

Machine Learning Intern at MVDSS Services LLP, Dabra, Madhya Pradesh

Built a Next Best Item prediction pipeline improving recommendation accuracy by 15% and supporting retention potential. Conducted price elasticity simulations to help define pricing strategy with data-driven insights. Automated feature extraction in Python, reducing preprocessing time by 30%. Processed large-scale datasets on Databricks using PySpark, improving scalability and reducing job runtime by 20%.