Opening the field notes · Systems
Unfolding the system story.
live
Compare brand mentions and citations across repeated checks with AI providers.
A single AI answer is a weak signal. CiteTrail preserves repeated provider responses and citations so teams can inspect brand visibility over time.
My role
Ketrium Labs · co-founder and engineer
Contribution and scope
As co-founder and engineer, I owned the full-stack architecture across prompt orchestration, auditable model responses, multi-tenant workspaces, analytics, LLM integrations, asynchronous processing, authentication, cloud infrastructure, and production deployment.
Model integrations
6 engines · 5 providers
Six API-backed configurations cover five provider families: OpenAI, Perplexity Sonar, Gemini, Azure OpenAI, and Claude. One Gemini configuration uses limited Search grounding. CiteTrail queries provider APIs, not consumer apps such as ChatGPT, Microsoft Copilot, or Google AI Overviews.
Visibility signals
4 lenses
mentions, recommendations, competitors, and citations/sources
Evidence
Inspectable
stored responses and citations stay connected to each run
Capability map
Related capabilities, grouped for exploration. This is a capability index, not a diagram of internal services or execution order.
A single model answer is a weak signal. Teams need repeated observations across provider APIs, with the response and citation kept available for inspection.
Tradeoffs suggested by the documented capabilities.
Repeated runs expose variation in mentions and recommendations instead of treating one response as a lasting truth.
Tradeoff: More useful comparisons require orchestration and careful handling of asynchronous results.
A visibility trend is more useful when a reader can inspect the answer and sources that produced it.
Tradeoff: Stored evidence adds data-lifecycle and workspace-isolation responsibilities.
Tools used in this system.
Next.js
React
TypeScript
Go
Fiber
GORM
PostgreSQL
Redis
Google Cloud Run