Software Engineer · Inflight (first engineering hire)
Jun 2025 — Present
Remote
- Built an offline LLM evaluation harness using LLM-as-judge across five quality dimensions; drove seven tuning iterations over a 36-PR, 33-repo corpus, lifting design guides 4.06 → 4.61 and code guides 4.22 → 4.55.
- Instrumented LLM analytics across the Claude proxy and guide agents — multi-turn traces plus a per-user token ledger with effective-dated pricing and org-level spend attribution.
- Shipped an AI workspace assistant with tool calling, streaming, and hybrid vector + keyword retrieval behind a feature-flagged rollout.
- Built agent-driven visual PR review on playwright-mcp with agentic preview spin-up across six language ecosystems; published the company Claude Code plugin and MCP server.
- Sole engineer alongside the CTO across a TypeScript/Hono/Bun API, a React web app, and a SwiftUI macOS client.