AI Engineering Intern at Nadeera Technologies
- Built the agent behind Meridian, an HR policy assistant: it chains tool calls (document search, SQL lookups, a calculator) in a hand-written OpenRouter loop and tailors answers to the employee asking.
- Tagged each answer sentence with its source, added a verifier that removes sentences it cannot trace, and resolved conflicting policies by effective date instead of silently picking one.
- Built the scorer for a 60-question eval that needs no LLM judge (79% groundedness, 86% refusal accuracy, 0.72 tool-selection F1) and backed the system with 450+ unit tests.