Helping enterprises break the linear - merging technical mastery with strategic insight for mission-critical solutions. #AIOrchestration #SovereignCloud
Santa Clara, CA
Joined July 2017
- Tweets908
- Following167
- Followers272
- Likes286
With AI coding agents, you no longer need to touch every line yourself, but supervision stays. Unlike a framework like #Spring, agent output differs on every run and lands in your own repo.
Albert Bernat on what a #ClaudeCode workflow taught his team: grapeup.com/blog/ai-developm…
Who uses AI in your company, which models, and what does it cost? If that takes over a minute, you have a visibility problem.
Aiboostr puts every model and agent behind one gateway. You set who uses which AI, within what limits, and see who spends what.
grapeup.com/aiboostr
🤖 Made with AI
Off-the-shelf airline #CrewPlanning tools do their own job very well. The trouble starts between them: a what-if, the annual report, a forecast a year ahead. Planners export the data and rebuild it in spreadsheets.
Planners: what still ends up in Excel for you?
In many companies, AI has outgrown a few API keys. It needs structure.
Aiboostr puts one gateway between your teams and every model: rules and token budgets set once per team.
Who owns token budgets in your company: IT, finance or each team?
grapeup.com/aiboostr
Two crew planning teams can both be right and still disagree.
“Two different teams can have two different formulas for the same concept,” says our business analyst Jarosław Gruszka. So planners pad standby across the whole airline.
How do you set standby today?nie po
Most IT budgets assume cost follows usage. #AIAgents break that: the same agent on the same task can produce token bills up to 30× apart.
For #FinOps, the realistic goal is bounding spend, not forecasting it.
By Roman Swoszowski: grapeup.com/blog/ai-agent-co…
In legacy modernization, AI code is only as good as your ability to audit it.
G.Tx tracks every run node by node and has a second model grade each agent's work. Here it rebuilds COBOL in C#.
What would you check before signing off?
100+ AI solutions in under two years, with the same #DataEngineering team.
That's what a premium automotive manufacturer got after we automated their data platform and gave analysts no-code self-service. Weeks-long queries now take minutes.
Case study: grapeup.com/case-studies/how…
The #EUAIAct in five words: prove your AI is under control.
That proof is evidence, and evidence can't be produced after the fact. Either you captured it along the way, or you didn't.
This is the #AIGovernance gap most teams still haven't closed.
Most teams test a private #LLM on the wrong question: can it match a frontier model everywhere? It only has to be good at the few jobs you'll give it. There, the setup around it counts most.
Four rules for a #SovereignAI model test, by Roman Swoszowski: grapeup.com/blog/private-llm…
Your customers can now tell you to share their #IoT device data with another company.
Under the #EUDataAct, you have to hand it over and prove what each customer agreed to.
Watch Databoostr handle one request in 50 seconds. databoostr.io
"It can't leave our environment."
In regulated industries, that line kills more AI projects than budgets do. The best use cases sit on data that #DataResidency rules lock down.
#PrivateAI flips it: the model comes to the data. See it in Aiboostr ↓ grapeup.com/aiboostr
#AIAgents break cost forecasts: same agent, same task, token bills that vary by up to 30×. For #FinOps, the goal is bounding spend, not forecasting it - a hard cap per run at the #LLM gateway, plus traces showing which step grew.
grapeup.com/blog/ai-agent-co…
The EU Data Act requires manufacturers to give users direct access to their product data.
Today that means a support ticket, an engineer with database access, and a CSV a week later.
Here is what real direct access looks like.
databoostr.io
"Rewrite just this one part" sounds like the safe option: less code, smaller scope, lower risk. Then you try to draw the line. In a tightly coupled ten-language platform, the part the business talks about has no boundary in the code. #LegacyModernization #SoftwareArchitecture
Only then is there something to cut. 48 epics landed in scope, covering 251 user stories - the part to be rewritten in C#, with an estimate you can actually defend.
That trace is the whole point: an estimate you can follow down to the code it covers.
Full write-up by Sylwia Cieśla: grapeup.com/blog/how-to-scop… #ApplicationModernization #CSharp