Case study · 01
Asoma
An AI engine that measures what ChatGPT says about a brand
Asoma is a SaaS that tracks when ChatGPT, Gemini, Claude, Grok and Meta AI recommend a brand or its competitors, explains why, and turns it into actions. I designed and built its AI engine and analysis layer.
- My role
- AI engine design and development, at Prex Technologies
- Stack
- Next.js · Multi-provider LLM APIs · AI agents · Data pipelines
Results
- 5AI assistants measured: ChatGPT, Gemini, Claude, Grok and Meta AI
- 8pipeline stages, from the question to verifying the result
- 4specialised agents that set up, analyse and recommend
- 0–100visibility score, plus share of voice, position and citations
The challenge
People now ask AI assistants what to buy, and brands have no idea what those assistants answer. The answers change by model, by week and by how the question is phrased, so a single manual check proves nothing. The product needed repeatable, auditable measurement across five different models, and a way to turn raw answers into decisions a marketing team can act on.
What I built
- A measurement engine that sends the same versioned questions to every model each week and stores the exact model, method, parameters, full unedited answer, native citations and run status, including retries and failures.
- Scoring that turns answers into a 0–100 visibility score, share of voice against competitors, position in the answer and whether the brand's own site is cited.
- Source intelligence that shows which pages the assistants rely on, so teams know where to act.
- A team of agents (coordination, setup, analysis and recommendations) that closes the loop: measure, understand, act, measure again, with before/after experiments.