Project / B2C product with applied AI
A post-payment async pipeline that prepares interviews without guessing.
A B2C product in production with multi-LLM generation, structured output, schema validation, fallback, retries, telemetry and human review.
The engineering problem
Preparing for an interview means reading a résumé against a job posting and producing consistent material. That cannot depend on a single model call or on unvalidated output, especially after the user has paid.
What was built
A post-payment asynchronous pipeline with Next.js, Inngest, Turso and Cloudflare R2. Multiple LLMs produce structured JSON validated by schema, with provider fallback, timeouts, retries, rate limiting and token and cost telemetry.
Quality controls
Gates prevent fabrication and inconsistency, and when automated quality criteria are not met, the item is routed to human review. Models are benchmarked by quality, latency and cost before selection.
Evidence boundary
The product is live and publicly accessible. The case does not promise interview approval and does not publish usage metrics.
WhatsApp / (85) 98596-3329