Chris Huang

– Took ownership of Finn and scaled it from an internal prototype to a production AI agent; now automates 97% of customer support interactions, handling millions of queries and freeing up 80% of human support capacity. – Built real-time speech-to-speech translation across 38 languages using a pipeline of OpenAI Whisper (ASR), fine-tuned NLLB models (NMT), and Coqui TTS; reduced average resolution time for non-English queries by 70%. – Authored architecture decision records (ADRs) for the Finn multi-language speech pipeline and facilitated design reviews across the engineering org, aligning teams on latency budgets, model selection, and GDPR data-compliance requirements. – Led a major model upgrade in H1 2026, migrating from a generic LLM to a domain-adapted model (fine-tuned Llama-3-8B using PyTorch and HuggingFace) with continuous feedback loops and A/B testing; improved user satisfaction from 83% to 90%. – Designed the agent's microservices (Go) on AWS EKS, ensuring 99.9% uptime and p95 latency under 200ms for all endpoints. – Built a GDPR-compliant data pipeline for user audio and transcriptions, implementing automatic data minimisation and right-to-erasure hooks; passed two external security audits without findings. – Established AI-augmented development practices (Cursor with bunq-specific rules, Claude Code) across the Finn team, accelerating feature delivery by 30% and reducing defect rates. – Mentored two mid-level engineers, helping both gain ownership of major Finn sub-components and achieve senior promotion readiness.

– Led the rewrite of the multi-step clinic onboarding flow (React + GraphQL), adding draft persistence and email recovery, reducing user drop-off by 22% and improving completion rate for new clinics. – Built a review verification system (Ruby on Rails + Sidekiq) that cross-checks patient email domains against appointment records, flagging an average of 200 fake reviews per month and preserving platform trust. – Designed a token-bucket rate limiter (Redis) for the public booking API, preventing three abuse incidents in the first six months and ensuring 99.9% availability during peak traffic. – Across a team of 5 backend engineers, restructured the GitLab CI pipeline by implementing layer-aware caching and parallel test sharding, reducing deployment time from 15 to 5 minutes and enabling twice-daily production releases without after-hours scheduling. – Established repo-specific Cursor rules for Ruby style consistency, reducing PR review time by 25% across the backend team of five engineers. – Partnered with compliance and legal to design a cryptographically-signed audit trail (RoR + PostgreSQL) covering all 23 patient-data endpoints, incorporating automatic data-retention policies and right-to-erasure hooks; passed two consecutive external GDPR audits with zero observations, reducing compliance risk for a platform handling 2M+ patient records. – Mentored a junior developer through promotion to mid-level, including code review and career guidance.

– Owned the migration of 50+ legacy cron jobs to Google's internal workflow engine, designing migration tooling and rollback scripts, reducing weekly maintenance overhead by 80% and eliminating three recurring production alerts. – Built a React + Java dashboard for ad campaign metrics, used by 150+ sales reps; reduced support tickets related to data discrepancies by 25% within three months. – Refactored a high-throughput log aggregation service (C++), implementing batching and compression; reduced p99 write latency from 120ms to 45ms and saved 30% in storage costs. – Analysed test reliability patterns across 50+ services, systematically removed 40% of flaky integration tests and implemented a quarantine suite, cutting CI pipeline runtime by 8 minutes while reducing false-negative triage time by 15 engineer-hours per week. – Drove adoption of automated code review tooling across the team, cutting PR nitpick comments by 30% and shortening review cycles by one day on average. – Partnered with product teams to prioritise features that directly improved ad delivery latency, contributing to a 15% reduction in 95th percentile latency. – Mentored two junior engineers through their first full development cycles; both were promoted to L3 within 12 months.