The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
I rebuilt my Pecauva case study so you can actually use it: click through the architecture, run the research matrix, and watch the sandbox recover from a failed run. Visit Pecauva
The sandbox retry retries from the saved checkpoint that it already failed in and it actually reviews itself against the system prompt and the rules if the document is not matching any of them it will start over.
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Related posts
Your own secure AI workforce, built on your own data.
Most SMEs pay for one expensive, generalist AI that knows nothing about their business. We're flipping that: a small, focused AI assistant for each department (HR, Accounts, Marketing, Sales, and Operations), each trained on your own business data.
How we build it:
The result is lower cost, higher accuracy, and an AI that actually understands how your business runs.
We're piloting this internally at Indigloo first. If you run an SME and want to see a department assistant in action, DM me. 🚀
For the internal pilot, I'd keep a held-out question set per department and compare RAG alone with RAG plus fine-tuning. Which department do you expect to benefit most from tuning, and what errors would you want it to reduce?
I built an AI support-ticket router on Replit, then tested it against the real model. Three things broke that no mock would have caught.
A silent failure. The first run sent 33 of 40 tickets to "needs a person". The newer Claude model rejects the temperature setting, and my code swallowed the error, so every call quietly failed. Now every failed call is logged.
The model thinks before it answers. With a tight token limit, some tickets got no answer at all and some draft replies stopped mid-sentence. Both have room now, and a cut-off draft is never saved.
One incident looked like four bugs. Four Safari checkout failures each got a different issue key, so the "3+ tickets, same problem" alert never fired. Sorting tickets in arrival order and asking for keys that name the problem rather than the customer turned them into one flagged incident.
If you're adding an LLM feature to your product, test it on real inputs early. The failures are rarely where you expect.
Live demo: https://sorta--krishnanbhupath.replit.app
what do you feel as an designer or user, which one feels more graceful ?
3 voted
75%
1 voted
25%
4 votes
Closed
