Skip to main content
veles

// Custom AI Development

Custom AI development for businesses that ship

End-to-end AI builds for SMB owners, SaaS founders, and operators across Europe. We do the architecture, the engineering, and the AI tuning ourselves. Production-quality systems in 2-8 weeks. The code is yours, the data is yours, the leverage is yours.

What custom AI development means at Veles

Custom AI development is the umbrella for every project we ship. Some are automation. Some are conversational agents. Some are analytics dashboards. Some are AI features built into someone else's SaaS product. The common thread is that every project is bespoke, designed around your real workflow, and built by a small senior team that owns the work end to end.

We do not resell a no-code stack with a markup. We do not sub-contract to a body shop. We do not ship a generic chatbot with your logo on it. We write real code, run real reviews, and ship production-grade AI that you can trust with revenue-critical workflows.

What custom AI development is not, at Veles:

  • It is not ML model training from scratch. We use the best frontier models (OpenAI, Anthropic, Google) and tune the behavior with prompt engineering, retrieval, fine-tuning, and tool use. If your project genuinely needs a custom-trained model, we will tell you on the discovery call and recommend someone who specializes in that
  • It is not a multi-year enterprise consulting engagement. Our projects are 2-8 weeks, fixed scope, fixed price, owned by you at the end
  • It is not "AI strategy decks". We build working software. The deliverable is code in production, not a 40-slide PowerPoint

Where we specialize

Custom AI development at Veles routes into three specialized service tracks. Most projects land in one of them; some span two or three.

  • [AI Automation](/services/ai-automation) — workflow systems that handle order intake, document processing, scheduling, lead qualification, and other repetitive operational work across the tools you already use. Best fit for SMB owners drowning in operational overhead
  • [AI Agents](/services/ai-agents) — conversational agents for WhatsApp, Instagram DMs, email, and your own channels. They read inbound messages, draft responses, book or order what they can, and route the rest to humans with context. Best fit for high-message-volume businesses
  • [AI Analytics Dashboards](/services/ai-analytics) — custom dashboards that pull from messy sources, apply AI where it actually helps (scoring, anomaly detection, segmentation, NL summaries), and surface the metrics that drive this week's decision. Best fit for founders who need one screen that tells them what to do today

Your project may not fit cleanly into one box. That is fine. We scope each engagement around the actual problem, not around a service line.

The four-stage delivery process

1. Discovery. A 30-minute call. Free, no commitment. You walk us through the workflow, the volume, and what good looks like. We listen and write up what we heard. By the end of the call you know whether Veles is the right partner. If we are not, we will tell you so on the call.

2. Scope and estimate. We map the project into milestones, identify the integrations and the hard parts, and write a concrete fixed-price estimate within 2-5 working days. No vague "starting from" numbers. See our scoping methodology for how we get to a number we both trust.

3. Build. A small senior team works in weekly sprints. You get a live demo each week, real working software, not slides. You can ask for changes mid-build without paperwork. The system goes through internal review before it ever touches a real customer.

4. Launch and operate. We deploy and monitor for two weeks, tuning the system on real traffic. After that, hand off the code with documentation or stay on a monthly retainer for ongoing operate-and-iterate work. Either way, the code is yours, documented, and built to be maintained.

Who we build for

Three ICP buckets cover most of our work, and we know each one's pain in detail because we have shipped it.

  • SMB owners across Europe growing past the point where the founder can keep doing all the operational work personally. We build the AI layer that absorbs the repetitive parts, so the owner spends time on the work that actually requires them. Mobilni Market is exactly this
  • SaaS founders who need an AI feature shipped into their product without hiring a full ML team. We build the AI layer, integrate with the existing stack, and hand off documented code. People Up is exactly this
  • Operators in e-commerce, B2B, recruitment, and local services who are drowning in messages, leads, or operational decisions. We build the agent or dashboard that turns the noise into a focused work queue. Kacimbas is exactly this

If your problem rhymes with any of these, the discovery call will tell us in 30 minutes.

Why custom over off-the-shelf, in-house, or other agencies

vs. off-the-shelf AI products (Zapier, Make, Intercom, Drift, Looker Studio): Off-the-shelf is great for stable, narrowly-defined needs. It falls apart on judgement-based decisions, non-SaaS data sources, high-volume processing, WhatsApp Business at scale, and anything role-specific. Custom AI gives you the leverage off-the-shelf cannot.

vs. hiring an in-house AI team: Hiring a senior AI engineer in Europe takes 4-6 months and costs €100k-150k loaded, before the first project ships. Veles ships a working system in 2-8 weeks for a fraction of that. Most clients use Veles to validate that AI is worth investing in before they hire.

vs. other agencies: We do not resell no-code with a markup. We do not sub-contract to body shops. We do not push enterprise consulting engagements on small businesses. The team that scopes your project is the team that builds it and the team that operates it post-launch. When you have a question six months in, the person who answers is the person who shipped the code.

vs. AI strategy consultants: They ship slide decks. We ship code in production. Pick the one that matches what your business actually needs.

// Proof from production

Already running in production

// FAQ

Frequently asked

What kinds of custom AI development projects do you take on?
Most projects fit into one of our three specialized tracks: AI automation (workflow systems), AI agents (conversational), or AI analytics (dashboards with AI features). Plus custom AI features built into someone else's SaaS product. We do not do ML model training from scratch or multi-year enterprise consulting. If your project does not fit, we will tell you on the discovery call and recommend someone who specializes in it.
Do you train custom AI models?
Rarely from scratch. For most business problems the best frontier models (OpenAI, Anthropic, Google) plus prompt engineering, retrieval-augmented generation, tool use, and targeted fine-tuning solve the problem faster and cheaper than a custom-trained model. If your project genuinely needs a custom-trained model (rare, usually research-grade work), we will recommend a partner who specializes in that.
How is custom AI development different from buying off-the-shelf AI tools?
Off-the-shelf tools (chatbot platforms, BI tools, no-code automation) are great when your needs are stable, narrowly defined, and match the tool's default feature set. They fall apart on judgement-based decisions, non-SaaS data sources, high-volume processing, compliance-sensitive work, and anything that needs to be role-specific. Custom AI development covers the gap where off-the-shelf either does not work or costs more in subscriptions over 18 months than a custom build cost upfront.
How long does a typical custom AI development project take?
Most projects we ship are 2-8 weeks end-to-end. A focused single-channel build (one input source, light integration, no custom dashboard) lands in 2-3 weeks. A multi-channel system with custom integrations and an operate-and-iterate retainer runs 6-8 weeks. We give you a concrete timeline at the end of Discovery, before any code is written.
What does a Veles project typically cost?
Depends on scope. A focused single-channel AI agent or single-source analytics dashboard is on one end of our range. A multi-channel system with custom integrations, on-premise inference, and an operate-and-iterate retainer is on the other end. After a 30-minute discovery call you get a real fixed-price estimate. We do not publish tiered pricing because every custom AI build varies too widely. See our scoping methodology.
Do you work with companies outside Serbia?
Yes. About half our work is cross-border. We are based in Belgrade and ship to clients across Europe (Germany, Netherlands, UK, Nordics, France), and occasionally further. We work remotely with weekly video demos and Slack/email async between. Time zone overlap with EU is excellent; with North America the morning hours work fine for handoffs.
Who actually does the engineering?
A small senior team. The same engineers who scope the project also build it and operate it post-launch. We do not "sell senior, deliver junior." We do not sub-contract. When you have a question six months after launch, the person who answers is the person who shipped the code.
What happens to the code after we launch?
You own it. Full code, documentation, runbook, and any deployment scripts. If you have an engineer on the team, they can take over operations entirely. If you do not, we are available for ongoing operate-and-iterate work at a predictable monthly rate. Most clients start with hand-off and switch to retainer after 1-2 months if the system needs ongoing care.

How we estimate

No tiered pricing because every custom AI development project varies. After Discovery we map the project into milestones, identify the integrations and the hard parts, and give you a concrete fixed-price estimate within 2-5 working days. No surprises during build, no vague "starting from" numbers in the proposal.

See our scoping methodology →

Talk to us about your AI project

Bring the workflow, the volume, and what good looks like. 30 minutes is enough to know if custom AI development is the right fit for what you are building. We will tell you straight if it is not.

We use cookies to analyze site traffic and improve your experience. Privacy Policy