AI&workflowautomation
AI workflows that replace manual handoffs — billing, onboarding, triage, content ops. Built from Novi Sad, Serbia for studios in the Balkans and EU. Frontier models when needed, self-hosted Llama or Mistral when tokens matter.
- From
- €4,500
- Timeline
- 4–8 weeks
- Tech
- n8nTemporalInngestClaude APIOpenAIOllama (self-hosted)Llama 3MistralLangGraph
What you get
What you get
- 01Workflow inventory and ROI model
- 02n8n, Temporal or Inngest orchestration layer
- 03LLM steps — Claude / GPT or self-hosted Llama 3 / Mistral
- 04Evaluation harness (golden sets, regression tests)
- 05Human-in-the-loop review queues where needed
- 06Cost dashboard: tokens, GPU hours, flat rate compared
How we deliver
How we deliver
Scope
Pick 2–3 workflows with clear ROI. We say no to the ones that don't pay back.
Prototype
A working end-to-end slice in week two, even if ugly.
Harden
Evals, guardrails, monitoring. Offline Llama / Mistral on your GPU if you want zero token cost.
Hand over
Your team can edit, retrain, and extend without calling us.
Related work
Related work
HK Vojvodina — one system for a hockey school
The whole club in one place: players, parents, coaches, training, payments, competitions, stats. Plus the public website.
that actually
ships.
OHM Agency — engineering half of a creative studio
Long-running partnership with a Belgrade creative studio. We do the engineering, the AI work, and the internal tools that keep everything running.
Journal
Keep reading
- teardown · 6 min
Self-hosted Llama vs Claude API: cost breakdown
When a token bill is the problem, and when a GPU is. One month of real numbers from a working agency.
- essay · 5 min
Engineering as the other half of a creative studio
Why the classic agency design-to-dev handoff fails, and what we do differently with OHM.
Ready when you are
Havesomethingtobuild?
Tell us what you're working on. We read every message and reply within one business day — with a real opinion and a rough number.