Founder & Backend Engineer
Icebreakr · Jan 2026 – Present
- Designed a plug-and-play provider architecture in Go where LinkedIn data sources (LinkdAPI, Apify) and LLM backends (Anthropic Claude, OpenAl, Gemini) sit behind interfaces with automatic fallback chains, swappable from a single config line.
- Hardened the LLM layer against prompt injection untrusted profile data is sanitized and wrapped in EXTERNAL_DATA delimiters so the model treats it as data, never instructions - with a typed-error retry loop that re-prompts on banned-phrase or structure failures.
- Built a Go eval harness with deterministic auto-checks plus an LLM-as-judge scoring four subjective axes, using baseline / diff modes to catch prompt regressions before merge.
- Built the Go backend (Gin, pgx, PostgreSQL, Goose) with errgroup-orchestrated parallel fetches behind a 75s pipeline timeout, per-IP and per-user rate limiting, daily usage caps, and structured slog logging.
- Built a Manifest V3 Chrome Extension in TypeScript + Vite using a closed Shadow DOM overlay and cross-origin background messaging, with no remote scripts as required by Chrome's MV3 policy.
- Implemented WorkOS OAuth / SSO with HttpOnly session cookies and a 30-second signed handoff that bridges website sign-in to the Chrome extension without exposing tokens to JavaScript.
- Designed a three-tier profile freshness system (manual refresh, soft-banner and silent auto-refresh windows, and a synchronous force-refresh) so generated outreach never references stale LinkedIn data.


