AI Product & Research Intern · Sentient Labs
May 2025 – Aug 2025 · New York City, NY (Hybrid)
- Built an internal evaluation suite of 100+ LLM prompts with automated scoring for tool-call correctness and safety-violation rate.
- Improved feedback-to-ticket latency by 20% via regression harness and iteration notes; shipped 3 tool integrations with engineers.
- Built a real-time crypto signals dashboard (2–3k lines backend TS, 1k lines frontend React) with latency under 2s for exchange data.
- Prototyped 'Dobby Mode' agent (300-line Streamlit demo) using Dolphin Mistral; enforced JSON schema to reduce hallucinations.