July 2025 – Present
Prompt Engineer – Prompt Factory
Cognizant · Hyderabad, India
- Built reusable components for the Prompt Factory auto-rater: policy-rule routing, RAG for policy context, and schema-defined JSON outputs for human review.
- Consolidated about 320 policy checks into 129 atomic rules (roughly 60% fewer). Wrote few-shot prompts with explicit flag/no-flag criteria for shared and workflow-specific rules.
- Automated concurrent LLM batch evaluation in Python. Reached 99% accuracy on selected categories against human-reviewed labels and refined prompts using per-rule error analysis.
- Integrated multimodal Gemini into an auto-rater that evaluated 50,000+ assets across 12+ workflows, reducing manual review time. For YouTube ads, it compared original and enhanced video to flag clipped text, audio cuts, and blur.