AI agents that act,
and ask first.

I'm Ergini, an AI engineer. I build agents that work inside the tools you already run, your inbox, CRM, calendar or ticketing, and hand the risky step to a person. I scope, build and ship it myself, on CET.

A free 30-minute call, on Google Meet

Prefer email? ergin@ergini.com. I reply within hours.

What I build

  • Task agents

    One named job, done end to end: qualify a lead, triage a ticket, draft the reply, book the slot. Five to ten tools, an eval set, and a person confirming the step that matters.

  • Automations between your systems

    The copy-paste between your CRM, inbox, spreadsheets and ERP, automated end to end. Code does the counting, the model does the reading, a person confirms.

  • Voice and chat agents

    Agents that answer, qualify and book around the clock, and hand off to a person with the whole conversation attached.

  • Agents inside your product

    An agent your own customers use, built on OpenAI, Claude or open models, with evals, cost ceilings and observability from day one.

Built so far

  • Solo build

    Caldra AI

    Scheduling agent for the calendar and inbox

  • Solo build

    VC Automation

    Lead research and outreach

  • Solo build

    Xandidate

    Candidate screening and ranking

All work

After the call

  1. Scope

    A written plan and a price within three days.

  2. Build

    Working software every Friday.

  3. Launch

    Deployed, documented and yours.

A single-purpose agent usually lands between $12K - $30K and ships in 2 to 5 weeks; a multi-agent system runs $25K - $60K. You get a written plan and a price within three days of the call.

Questions

How is an agent different from a chatbot?
A chatbot answers. An agent acts: it calls your tools, your database and outside services in a loop until the job is done. I build both, and I will tell you when a plain workflow is the cheaper answer.
How do you keep it from doing something wrong?
Tight tool schemas, strict output validation, step and cost budgets per run, and a human approval step on anything irreversible. The agent proposes; a person confirms the risky part.
Which models and stack?
OpenAI, Anthropic Claude or open models, whichever fits the job and your data rules. TypeScript with the Vercel AI SDK for most stacks, Python where your team lives there. Evals and cost logging are part of every build.
Who does the work?
I do. No account manager, no handoff to a junior. The person on your first call writes your code, and you own all of it.
How long does it take?
Agents and automations take two to eight weeks. You see a working demo every Friday.
Where are you based?
Pristina, Kosovo, on Central European Time. I work with clients across Europe and North America, on their hours.

Book the free 30-minute call