Wolven Tech: Rust consulting. AllSource Prime: durable AI memory → https://teapawt.net/t.co/CfdDxoKhs5 Longhand: AI workflows & training → https://teapawt.net/t.co/Rbml5TOtFF
London, United Kingdom
Joined January 2014
- Tweets4.6K
- Following2.8K
- Followers1.1K
- Likes9.3K
Pinned Tweet
1,000 followers. Thank you.
Marking it by opening something up: I am taking on fractional full stack AI developer work through Wolven Tech Ltd.
Five spots. One is already taken.
How it works below.
A successful Claude request can carry a profile ID the provider never looked up. Trace one customer through a retry and a batch before adding a provider route. 1/5
🤖 Made with AI
ALT A sculpted wolf carries a red identity ribbon along a branching path towards a silver checkpoint. Caption: Keep tenant attribution.
Trace a tenant through retries, batches and supported fallback routes. Compare intended identity with the outgoing profile and the platform’s checks. Attribution loss is a proposed application failure test, not a reported provider defect. 4/5
Which path would you inspect first for lost attribution: retry, batch or fallback? Reuse provider enforcement where it supplies the guarantee you need.
platform.claude.com/docs/en/…
5/5
The checkpoint hash matches. Activation precision may have changed underneath it. Before approving an inference optimisation, check which arithmetic ran. 1/5
🤖 Made with AI
ALT A checkpoint cube sits above two different precision apertures and a mechanical selector. Caption: Record activation precision.
Retain weight format, activation format, hardware and kernel configuration. Check task outcomes on release cases and keep the qualified configuration available for rollback. A checkpoint hash leaves those execution choices unanswered. 4/5
Which task outcome would make your team reject a faster inference path? Agree that boundary before inspecting latency.
vllm.ai/blog/2026-10-21-humm…
5/5
An engineer asks for more agent budget. What will the next attempt establish? Ask for the unresolved question and the observation that would change the decision to continue. 1/6
🤖 Made with AI
ALT A silver mechanical guardian examines a work card before opening a resource gate. Caption: Explain the next attempt.
Publish criteria beforehand. Review reasoning and responses to evidence across different tasks, with peer discussion. Tool spend and a lucky model answer cannot supply that judgement. This is unvalidated coaching, not an employment rubric. 5/6
What evidence would make you approve another attempt after an agent run stalls?
support.claude.com/en/articl…
support.claude.com/en/articl…
6/6
Requests keep arriving while new workers prepare. A brief scaling pause tells you little about that wait. Ask when the added workers take useful load. 1/5
🤖 Made with AI
ALT Mechanical support assembly prepares an unconnected segment beside a working bridge. Caption: Measure capacity readiness.
Measure trigger-to-readiness, serving interruption and queue growth separately. Repeat under the cache and storage conditions your deployment can encounter. Review reserve capacity and scaling thresholds against that evidence. 4/5
Spare capacity costs money. Does your dashboard show trigger-to-readiness? This proposed test promises no universal saving.
vllm.ai/blog/2026-10-19-elas…
5/5
