May 2023 – Present · 3 yrs 1 mo

• Led architectural redesign of AI-powered sales analytics platform, collaborating cross-functionally with PMs and PMMs to blend traditional software engineering with LLM capabilities, preventing hallucinations and ensuring deterministic business logic application for 100+ sales team users.
• Architected and implemented a hybrid SQL generation system using SQLGlot, Pydantic, and FastAPI that parameterizes free-form LLM interpretations into validated models with code-enforced business logic, dramatically reducing prompt-related regressions while preserving LLM flexibility.
• Engineered anti-hallucination architecture that separates concerns between LLM natural language understanding and SQL generation, improving accuracy by moving business logic from brittle prompts to testable code with strong validation patterns.
• Redesigned testing infrastructure by containerizing Trino database locally with Docker, enabling 100x test coverage with deterministic results and reducing test execution time from 30-60 seconds to under 2 seconds.
• Built sophisticated test framework featuring programmatic table schema extraction, mock data generation based on dimension/fact patterns, SQL function normalization, and dependency injection for fully deterministic test outputs.
• Mentored senior engineers on Python, code organization, library selection, and architectural patterns for AI-integrated systems, driving adoption of best practices across the team.