Fullstack Engineer
axess Intelligence GmbHCologne, Germany
Owning the platform's surface area — API design, query plans, the deployment pipeline, and the AI features the product now leads with. Recent work has centered on shipping LLM surfaces (an in-app chat assistant, chat-to-report, evaluation) on the Vercel AI SDK and a type-safe tRPC backend, alongside the migration to App Router, Drizzle, and Neon Postgres that the rest of the team builds on.
- In-product AI chat assistant in production: multi-provider LLM (Claude, OpenAI, Gemini) on the Vercel AI SDK, a streaming UI, and a type-safe tRPC backend.
- Chat-to-report — editable reports generated from chat with per-block AI regeneration and live streaming across text, chart, and table blocks; built on the v1 RAG pipeline and confidence-scored feature extraction.
- LLM observability and evaluation on Braintrust: request tracing, a custom scorer wired into CI, and online/offline judge parity for answer quality.
- Pages-to-App-Router migration with 90+ raw-SQL endpoints refactored into parameterized Drizzle queries on tRPC — end-to-end type safety, ~50% less debugging time.
- MySQL to PostgreSQL on Neon serverless, with indexes and CTEs on the hottest queries: up to 60% faster reads.
- Enterprise onboarding via Azure Entra SSO with per-organization enforcement — JWT, RBAC, and invitation and password-reset flows on better-auth.
- Test suite (unit, integration, E2E) with Stryker mutation testing and an A SonarQube reliability rating: ~40% fewer post-deployment bugs.