Gaurav Sonawane

IIT Bombay M.Tech Graduate and AI/ML Engineer specializing in production GenAI systems — RAG pipelines, LLM fine-tuning, AI guardrails, and MLOps. Experienced in building and deploying scalable, observable AI platforms on AWS.

AI/ML Engineer at Esyasoft Technologies (2025 – Present)

Project: Esyasoft Copilot Bot

• Contributed to deploying a 15+ container microservices AI platform on AWS EC2 GPU instances with Docker Compose, Keycloak SSO, inter-service health monitoring, and NVIDIA CUDA passthrough for vLLM inference serving.
• Built a fault-tolerant 22-guard AI safety pipeline covering input/output filtering: jailbreak detection, PII masking, hallucination blocking, schema leak prevention, and SQL injection validation.
• Built a high-throughput Theft Detection NL2SQL microservice querying 117M+ daily consumption records, enabling non-technical analysts to detect tariff misuse and theft patterns conversationally.
• Engineered a scalable, low-latency RAG pipeline (LangGraph) with query rewriting, HyDE, multi-query decomposition, and LLM-based re-ranking, deployed as an asynchronous FastAPI microservice.
• Designed a custom observability platform with P99.9 latency tracking, Apdex scoring, GPU/CPU/RAM telemetry, and Prometheus + Grafana dashboards for real-time service monitoring.
• Implemented CI/CD pipelines using GitHub Actions for automated linting, testing, Docker image builds, and deployment workflows across the multi-service platform.

AI/ML Engineer at Esyasoft Technologies (2025 – Present)

Project: UK Gas Regulatory AI (Ofgem UNC Compliance)

• Implemented a graph-powered MOD impact assessment agent (LangGraph + Neo4j) traversing a 10-node knowledge graph (Clause → Rule → Tranche → Component → Participant), reducing ROM turnaround from 5–10 days to 1 hour.
• Built a hybrid retrieval engine combining dense vector search (Qdrant), sparse BM25, and Neo4j graph traversal fused via Reciprocal Rank Fusion with Cohere cross-encoder reranking, over 1,600+ pages of legal UNC documents.
• Architected a 3-layer output guard with retrieval score circuit breaker, low-confidence chunk filtering, and citation density enforcement, eliminating hallucinated regulatory advice in compliance-critical responses.

Business Analyst Intern at Finlatics (2024-01 – 2024-02)

• Transformed complex datasets into actionable insights and developed KPI dashboards using Excel / Power BI.

M.Tech in Energy Systems Engineering – Indian Institute of Technology, Bombay (2025)

B.Tech – Indian Institute of Technology, Bombay

XII

X