Bhargav Kommineni

About

Machine learning lab assistant at Eastern Illinois University who productionized a voice-enabled conversational AI system: redesigned it with RAG, LoRA fine-tuning (including text-to-SQL), and safety guardrails, then deployed a scalable API supporting ~1,000 daily queries. Also partnered with customer-facing teams during a BlueFi internship by building demos/APIs and accelerating releases via Terraform + AWS CI/CD automation.

Experience

Education

Key Strengths

• Took an LLM prototype to production using a RAG architecture with guardrails
• Fine-tuned LLMs with LoRA for text-to-SQL; achieved ~80% exact-match accuracy
• Operated and monitored a live assistant handling ~1,000 daily queries (latency, error rates, response times)
• Strong real-time troubleshooting of LLM pipelines via stage-by-stage logs/traces and component isolation (retrieval vs prompt vs model vs dependencies)
• Built end-to-end voice conversational system (STT + TTS) and deployed as a scalable API
• Uses user feedback loops (interaction logging + failure analysis) to iteratively improve retrieval and prompts
• Effective technical communicator: leads developer-focused demos with architecture + code/API workflows + scaling/debugging guidance
• Bridges product/customer-facing needs and engineering execution; built demos and improved deployment speed via Terraform/AWS CI/CD
• End-to-end ownership of full-stack feature (React + Node/Java + Postgres) including deployment/ops
• Built ~20 REST APIs for auth, menu, cart, and order management
• Performance optimization with measurable impact (FCP 3.4s → 1.6s via code-splitting/deferring)
• Production-minded DevOps: CI/CD automation, IaC on AWS, and rollback capability
• Observability and reliability practices (CloudWatch monitoring/alarms; tracking response time/error rate/server health)
• Designed maintainable microservices via layered architecture, shared packages, and strict TypeScript contracts
• Standardized APIs and improved team alignment using OpenAPI specs
• Built reliable financial data ingestion pipelines with idempotency, retries, reconciliation, and backfills