Intelligent Ticketing System
Ticket creation for call-center agents, with the category predicted automatically.
- 2 weeks
- to ship v1 to production
- v2
- Jev integration in progress
An automated panel that lets Link3's call-center agents create support tickets with the ticket category predicted for them. v1 shipped in two weeks using cosine-similarity search over embeddings and runs in production. v2, in progress, integrates Jev (TypeSafe AI), a non-generative decision model with calibrated confidence.
v1: cosine-similarity predictionJun 1 – Jun 15, 2026
Built in two weeks on architecture groundwork from my internship (Feb–May 2026). Predicts the ticket category by cosine-similarity search over embeddings. This is what runs in production today.
v2: Jev integrationSep 2026 – present
Integrating Jev (TypeSafe AI), a non-generative decision model that returns typed, structured decisions (for example a Choice for the ticket category) with calibrated confidence in a single pass. Not in production yet.
Why a non-generative model
Generative LLMs are too expensive for production at this volume, and small models aren't accurate enough. Jev is expected to give the best accuracy for the cost.
- Ownership
- Designed and built
- Roles
- Product owner, Full-stack, DevOps, AI engineer
- Stack
- Next.jsEmbeddingsCosine similarityJev (TypeSafe AI)GrafanaPrometheus