Angie Fathalla

About

AI/LLM engineer focused on production-grade agent systems in high-stakes workflows, including a tax modernization initiative that automated tax form generation for 50,000 users. Demonstrates strong depth in reliability engineering, evaluation loops, and safe orchestration, with quantified improvements including pass-through gains from 62% to 94%, 30% lower token cost, and accuracy improvement from 88% to 99%.

Key Strengths

• Shipped LLM agent for tax form generation at 50,000-user scale
• Designed production-grade reliability with circuit breakers and multi-level fallbacks
• Strong hybrid AI architecture using LLM orchestration with deterministic calculation engines
• Built predictable multi-step workflows with state machines and Redis-backed shared state
• Implemented checkpointing and recovery to resume jobs after infrastructure failures
• Created offline and online evaluation loops with regression testing and human-labeled failures
• Drove major reliability gains via map-reduce redesign
• Improved pass-through rate from 62% to 94%
• Reduced token cost per form by 30%
• Improved calculation accuracy from 88% to 99%
• Designed safe ERP/operational data integrations with idempotency and hard validation constraints
• Prevented financially risky quote errors by blocking bad actions and routing to manual review
• End-to-end ownership across backend, middleware, and web services for life insurance quoting
• Designed granular business error responses to improve user experience and troubleshooting
• Legacy-to-cloud modernization: offloaded mainframe tax-form data processing to AWS (S3 + Lambda + Python)
• Risk mitigation via atomic/event-driven Lambda executions to prevent batch-wide failures
• Strong stakeholder alignment through demos, Q&A, and deployment readiness communication
• Structured debugging workflow using sandbox testing, backtracing, and AI-generated test scenarios

Experience

Education

Skills

Languages