We build software that survives production.

A two-person engineering studio in Belgrade. Web and mobile products, AI features and the infrastructure behind them, designed, built and shipped end to end.

Start a project Book a 30-minute call

01 / What we do

Full-stack product engineering, from the first call to production. You talk to the two people who build it.

Products & platforms

Web and mobile, end to end: the product people use, the system behind it, the deploys, the monitoring and the documentation. One team from the first sketch to the first real users and the year after.

  • TypeScript
  • React & Next.js
  • React Native & Expo
  • Node.js
  • PostgreSQL
  • Supabase

AI features

LLM features, retrieval and agents inside real products, with the parts demos skip: evals, latency budgets, fallbacks, cost and failure handling. AI where it earns its keep, nowhere else.

  • Anthropic & OpenAI
  • Mastra
  • Vercel AI SDK
  • pgvector
  • Evals

Modernisation & rescue

Systems that already work and now strain: migrations, performance, reliability, a codebase your team can change. Including products built fast with AI tools that now have to hold.

  • Migrations
  • Performance
  • Security hardening
  • Code audits
  • CI & observability

GEO & MCP

The technical surface AI engines read: MCP servers and the apps that plug into ChatGPT and Claude, llms.txt, structured content and the pipelines that show where you are cited.

  • MCP servers
  • ChatGPT & Claude apps
  • llms.txt
  • Structured data
  • Citation tracking

02 / How we work

Progress you can click on

Every week ends with something you can open, click and send to someone else. No status decks. The demo is the report.

You talk to the people who build it

No account managers, no agency on top of an agency, no handoff to people you have never met. The two names on this page are the two people in the repository.

Yours from day one

Your stack where you will own it, your repository from the first week, your accounts in your name. Documented for the next engineer, or the next model, so nothing depends on us.

03 / Four phasesNo surprises

Discovery1 to 2 weeks

  • Stakeholder interviews
  • Technical audit and risk log
  • Scoped proposal and staged estimate

We interview the people who will use and run the thing, audit what exists, and pressure-test the scope. You leave with a real proposal: a number, a timeline and a definition of done we both sign.

Architecture1 week

  • System and data diagrams
  • Working spike of the riskiest path
  • Weekly milestone plan

We choose the stack, draft the system and ship a working spike of the riskiest part first. The plan stops being a plan and starts being a project, with the unknowns retired before they become surprises.

Build4 to 10 weeks

  • Working demos every week
  • Production-grade code, tests and evals
  • Shared channel, async first

We ship. A working demo on a regular cadence, a shared channel for the day to day, and a recorded walk-through for the people who do not live in it. If something is going to slip, you hear it before it slips.

Handoff1 week

  • Repository, infrastructure, secrets
  • Architecture docs and runbook
  • Workshop and 30 days of support

We leave you with the repository, the infrastructure, runbooks, architecture docs and a hands-on workshop with whoever picks it up. The first week on your own should feel set up to win.

04 / The studio

Small by design, hands-on by default.

Jakša Mališić

Jakša Mališić

Architecture, backend, AI, infrastructure

Nine years in the TypeScript, React and Node ecosystem. Senior AI engineer at a US property-management platform used by over a million landlords, building agent systems and the internal tools around them.

linkedin.com/in/jaksm
Aleksa Mališić

Aleksa Mališić

Frontend, design in code, mobile

Builds the part people touch: interfaces, mobile apps and the admin panels behind them, designed in code rather than handed over from a mockup. Has shipped fitness and booking products with accounts, roles and payments.

linkedin.com/in/aleksa-malisic
Founded
2026, Belgrade
Typical project
6 to 12 weeks
Availability
Open for new projects
At a time
One project

05 / FAQ

Answers before the kickoff

If yours is not here, send it. We would rather answer in writing than dance around it later.

01Who do you typically work with?
Founders building a first or second product, teams inside a growing company who need to ship a specific thing, organisations that need a platform built and handed over, and businesses moving into AI. We do well with people who know what they want the software to do, technical or not.
02How long is a typical engagement?
It depends on the project. Many engagements run a couple of months, scoped into clear stages; some are shorter Discovery audits, others are longer, continued work. We give you a realistic range after the first call.
03How do you price?
We bill hourly. After a short scoping conversation you get a total estimate, broken down by stage, so you can plan around a real number. If the scope changes, we flag it and re-estimate before any new work starts.
04Do you sign NDAs?
Yes, always. Send yours over before the first call, or use ours. We do not share what is not ours to share, which is also why this page shows no client work.
05Who owns the IP?
You do. Full assignment on signed deliverables, and the code lives in your repository from the first week.
06Can you start immediately?
Often, yes. We take one project at a time, so if we are mid-build we tell you the real start date on the first call rather than squeezing you in.
07What if we do not know what we want yet?
That is what Discovery is for. A one- to two-week phase that gives you a scoped proposal, a system sketch and a clear go or no-go. If you do not continue with the build, the artifacts are still yours.
08Our product was built with AI tools. Can you take it over?
Yes. A product that got real users through Cursor, Lovable, Bolt or Claude Code and then hit a wall on bugs, security or scale is a common starting point. We start with a short code audit, say honestly what is solid and what is not, then harden, refactor or rebuild the parts that need it. You keep what works.
09How is your GEO work different from an SEO agency adding “AEO” to its menu?
Most “AI SEO” is the same SEO with a new label. We build the technical surface AI engines actually read: MCP servers, llms.txt, structured content and the pipelines that track citations across ChatGPT, Claude, Perplexity and AI Overviews. We are engineers, not content marketers.

06 / Contact

Tell us what you are building. A few sentences is enough.

office@malisic.dev
Budget