Skip to content
42

Applied AI studio · Novi Sad

Superintelligence, applied.

We are an applied AI studio from Novi Sad. Engineers and testers who design, build, evaluate and ship AI systems that hold up in production.

Bring the question Tour the system

.si — SI, as in superintelligence. We build the applied kind.

The core

The core: architecture

42 is the answer to everything. It is only useful once you know the question.

Every system we build starts at the centre. Before a model is chosen or a prompt is written, we work out what the problem actually is, where the data lives, what failure costs, and what “done” looks like.

Then we design the system around it, and only then do we build. Everything else on this page orbits that order of work.

  1. Understand the problem
  2. Design the system
  3. Then build

Capability

Agentic Systems

AI that does the work, not just talks about it.

What we build

  • Orchestrators and tool-using agents that plan, call your systems and finish multi-step tasks
  • Human-in-the-loop checkpoints where a decision needs a person, and clear audit trails where it doesn’t
  • MCP servers that connect models to your tools, data and internal APIs

In orbit

  • Claude
  • GPT
  • Gemini
  • MCP
  • Function calling
  • Agent orchestration

Capability

Knowledge & RAG

Your documents, finally answering back.

What we build

  • Retrieval grounded in your own data, with answers that cite their sources
  • Hybrid search that combines meaning and keywords, so exact terms are never lost
  • Knowledge bases over documents, wikis and internal platforms, kept in sync as they change

In orbit

  • Embeddings
  • Vector search
  • Rerankers
  • Hybrid search
  • Document parsing / OCR

Capability

Assistants & Natural-Language Interfaces

Talk to your product.

What we build

  • Assistants for customers and staff that know your product and know when to hand over
  • Natural-language layers that turn a sentence into the right API call
  • Multilingual by design, including Serbian and other regional languages

In orbit

  • Intent routing
  • Guardrails
  • Streaming chat UI
  • Multilingual

Capability

Speech & Transcription

Speech to text that knows your vocabulary.

What we build

  • Transcription pipelines with domain-aware correction for specialist vocabulary: medical, legal, technical
  • Real-time streaming and batch processing from the same pipeline
  • Speaker separation, so every line is attributed to the right voice

In orbit

  • Soniox
  • Whisper
  • LLM correction pass
  • Diarization
  • Streaming STT

Capability

Generative Media

Video, images and voice. Generated, on-brand, at scale.

What we build

  • Production pipelines for generated video, images and voice
  • Consistent characters and style from the first frame to the last
  • Automated checks before anything ships, so a person reviews exceptions, not everything

In orbit

  • ComfyUI
  • Diffusion image / video
  • TTS & voice
  • Automated visual QA

Capability

Custom & Local Models

Models that live where your data does.

What we build

  • Fine-tuning on your data, for the tasks where a general model isn’t enough
  • Private, on-prem deployment when the cloud isn’t an option
  • Quantization and serving tuned for cheaper, faster inference

In orbit

  • LoRA / QLoRA
  • Open-weight models
  • vLLM
  • Ollama
  • Quantization

Capability

Evaluation & QA

AI tested like software, not vibe-checked.

What we build

  • Eval suites and regression tests for LLM systems, run on every change
  • Automated QA for generated output: text, images, audio and video
  • Monitoring after launch, so drift shows up on a dashboard before it shows up in a complaint

In orbit

  • Eval harnesses
  • LLM-as-judge
  • Regression suites
  • Observability

Capability

Product Engineering

The software around the AI.

What we build

  • Full-stack web and mobile apps that put the model in front of the people who use it
  • APIs and integrations into the systems you already run
  • Infrastructure, deployment and the unglamorous rest, because a model alone isn’t a product

In orbit

  • TypeScript
  • React / Next.js
  • Node.js
  • Mobile
  • Cloud & DevOps

Flight plan

How we work

A flight plan, not a pitch deck. Every engagement follows the same five stages.

  1. Discover

    Find the real question: the problem, the data, and what success is worth.

  2. Prototype

    Build the smallest version that answers it, on your data, not a demo set.

  3. Evaluate

    Measure it against tests you agree with before anyone trusts it.

  4. Ship

    Put it into production with the software, guardrails and handover it needs.

  5. Operate

    Monitor, retrain and improve, because models drift and products grow.

About

Engineers and testers.

42 Solutions is an applied AI studio from Novi Sad, Serbia.

We are a small team of engineers and testers, and we build AI the way good software gets built: designed first, tested properly, and run in production long after the launch.

Contact

Ask the question

We have the answer. Bring the question.

— 42 Solutions