Lakshmi Swathi Sreedhar

About

Built and deployed a production-grade, multi-agent Text-to-SQL assistant that lets non-technical stakeholders query large enterprise databases in natural language. Uses Pinecone-based schema retrieval + LLM reasoning (Gemini/Claude/GPT) with a dedicated validation agent (schema/syntax checks and safe dry runs) to reduce hallucinations and improve reliability, while optimizing latency and cost via async execution and embedding caching.

Experience

Education

Key Strengths

• Built and deployed a production Text-to-SQL assistant for non-technical users over enterprise databases
• Designed multi-agent LLM workflow with schema retrieval + SQL generation + validation for reliability
• Strong hallucination prevention via RAG over schema and past queries
• Reliability engineering: validation agent with schema checks, syntax checks, safe dry runs, and auto-correction
• Latency and cost optimization using prompt refinement, embedding caching, and async parallel execution
• Framework fluency for agentic orchestration (LangChain, LangGraph, DSPy) with emphasis on determinism and loop prevention
• Structured evaluation approach: upfront metrics, edge-case testing, end-to-end simulations, and post-launch monitoring
• Took a text-to-SQL LLM prototype to production for an enterprise customer by redesigning into a multi-agent system
• Reliability engineering for LLM workflows using guardrails, validation, and fallback logic
• Fast real-time diagnosis of agentic pipeline failures via reproduction, step isolation, and log/validator inspection
• Customer-driven iteration: weekly reviews of failures and converting them into re-prompting and new validation rules
• Strong technical demo/workshop delivery with interactive edge-case testing and failure analysis
• Effective pre-sales partnership: tailoring demos to customer workflows to drive adoption and close deals