vikhyath D

About

LLM/agent engineer who has shipped multiple autonomous, multi-step agents to production (document-to-SOP conversion, test generation, code generation) using a custom Python DAG orchestrator with persistent state, tool-calling permissions, and structured outputs (Pydantic/JSON Schema). Demonstrates strong production hardening practices—semantic contracts, golden-dataset prompt regression tests, circuit breakers, and multi-level monitoring—and delivered large productivity wins (34 hours of manual writing reduced to ~20 minutes review; ~15–20 engineering hours/week saved).

Key Strengths

• Shipped multiple production LLM agents (SOP conversion, test generation, code generation) coordinated via tool calling and structured validation
• Drove major efficiency gains: reduced SOP/document conversion from ~34 hours to ~20 minutes of human review; saved ~15–20 engineering hours/week
• Designed reliability safeguards for LLM failure modes (assertion checks against service registry; semantic contracts between stages)
• Built custom Python DAG orchestrator with persistent per-document state enabling crash recovery and resumability
• Implemented self-healing validation/retry loops that auto-resolved ~85% of malformed structured outputs
• Established prompt/model change regression testing using a golden dataset (50 representative documents) with similarity scoring to catch regressions pre-deploy
• Strong production ops discipline: observability, cost tracking, first-pass success metrics, circuit breakers based on rolling failure rates
• Improved agent reliability through root-cause analysis of logs (e.g., fixed broken imports by adding a directory lookup tool and enforcing its use)

Experience

Education

Skills

Languages

Certifications