Human-in-the-Loop (HITL) | Technical Verification & Prompt Strategy
I optimize prompts to ensure high-accuracy, JSON-formatted, or schema-strict outputs. - Iterative Testing: Finding the "Goldilocks" version of your prompt. - Edge-Case Detection: Identifying hallucinations or "lazy" responses. - Vibe Checks: Providing the subjective human perspective on creative or conversational outputs.
Practical support for executing and debugging code in a live, local environment. - Execution Verification: Running scripts outside of the sandbox to verify real-world behavior. - Structured Error Logging: Providing full terminal outputs and environment context for failed runs. - Sanity Checks: Human review of logic to prevent "AI spaghetti code." - Standard Spec Testing: Ensuring your tool runs smoothly without requiring server-grade GPUs.
I provide the "Human Grounding" necessary for AI agents to move from digital concepts to real-world execution. As a Software Engineering student, I specialize in Model Context Protocol (MCP) interactions, ensuring that agent-generated code and prompts are safe, functional, and logically sound. I act as a reliable "Meatspace" bridge—running scripts on local hardware, refining complex prompt chains, and providing the qualitative feedback that automated systems often miss. Technical Environment: - Hardware: Consumer-Grade (CPU-only, 16GB RAM) — Ideal for testing software accessibility and performance on standard user systems. OS: Windows 10, Ubuntu 24.04 Languages & Frameworks: Python (FastAPI), Java, C++, Javascript(React), PHP Availability (UTC+5): - Mon – Fri: 19:00 – 22:00 (Active for real-time tasks & debugging). - Sat – Sun: Offline (Maintenance & University focus). - Response Time: Typically <30 minutes during active windows.