Founding Full-Stack AI Engineer
0+
Concurrent Users
0%
Latency Cut
0
Platforms Shipped
- Designed and built COROS's 24/7 AI coaching platform from zero — a microservice backbone serving professionals, founders, and leaders with LLM-powered coaching conversations at scale.
- Cut coaching AI response latency from 27–32s to 6–10s via semantic search optimization, making real-time dialogue feel natural rather than transactional.
- Engineered a multi-type memory system tracking user moods, aims, and key relationships across seven life dimensions — with topic isolation and time-based decay to prevent cross-session context bleed.
- Built automatic LLM failover across OpenAI, DeepSeek, Together AI, and Llama — when the primary provider goes down, a backup activates with zero downtime and zero developer intervention.
- Produced a professional-grade ElevenLabs voice clone end-to-end — from studio recording pipeline to production deployment — delivering human-like speech quality inside the coaching experience.
- Implemented AI guardrails to detect and block harmful or inappropriate coaching advice before delivery, ensuring every response meets responsible AI standards.
- Built an anonymous coach evaluation engine surfacing session quality metrics and improvement insights — giving coaches visibility into their performance without exposing user data.
- Hardened the platform with security layers across auth, data access, and LLM API boundaries to protect sensitive coaching conversations.
- Shipped the COROS iOS + Android app end-to-end — owned the full CI/CD pipeline and App Store / Play Store release lifecycle from build to production.
- Slashed GCP spend with request-based instance scaling and per-endpoint resource budgets.
LLM OrchestrationRAGNext.jsNodeGCPPineconeElevenLabsReact NativeCI/CD
