Description
Design and extend production-grade LLM applications and agentic workflows using
NestJS, XState v5, and the OpenAI SDK — flows include RAG, intent detection,
clarification, fulfillment, escalation, tool-use, and human-in-the-loop state machines
– Build and maintain the conversation-machine substrate: guard/action registries, flow
validation (ajv), DB-driven flow configs, and design-time tooling in Epicenter admin
– Build and evolve the AI systems behind Epic Support Assistant (ESA), the
player-facing support chatbot, and Agent Support Assistant, the AI copilot used by
customer support agents
– Integrate with MCP servers (Model Context Protocol) for tool-use and agentic behaviors
– Evaluate, benchmark, and tune models across providers including OpenAI, Gemini,
Anthropic, and future providers; own model selection decisions balancing quality,
latency, throughput, reliability, and cost
– Troubleshoot production LLM issues including hallucinations, retrieval failures, prompt
regressions, model drift, token inefficiencies, latency bottlenecks, and provider outages
– Build resilience mechanisms: retries, fallback routing, caching, streaming, rate limiting,
and provider routing
– Instrument and tune model quality using Langfuse (tracing, evals, prompt
management), evaluation datasets, A/B testing, prompt versioning, and production
telemetry
– Manage async workloads via BullMQ and caching with Redis; PostgreSQL persistence
via Kysely
