Gautier Billard

AI case studies

Build. Observe. Refine.

Lessons from building Ariath. The problems encountered, the engineering decisions and what the observations actually support.

02 / Latency and TTFT

Optimising TTFT: reducing latency between the user and the model

From the frontend to inference, reducing round trips and preparation to bring the first visible token closer to provider latency.

Read the case study

≈2,500ms

of latency saved

Estimated total reduction: frontend, backend and switching to Azure.

01 / Prompt caching

Prompt caching: optimising AI agent costs and latency

Prefix stability, conversation persistence and context freshness: an engineering case study on inference costs and time to first token in Ariath.

Read the case study

73.0%

of input tokens served from cache in October 2026

13,527 included requests · multiple models and providers