Contact

AI systems

AI that's private, practical, and yours to keep.

Language models that work on your data, handed over at the end like everything else we build.

Tell us what data you have

How it works

Three shapes of AI project

Nearly every job is one of these, or two of them together. Pick one to see how the pieces fit.

RAG pipelineTwo lanes sharing a vector database. Ingestion: documents are parsed or read with OCR, split into chunks, embedded, and written to the vector database. Query: a user question is embedded, the closest chunks are retrieved and optionally reranked, then given to the language model together with the question, which answers with citations back to the source.IngestionQuery
Documents
Parse / OCR
Chunk
Embed
Vector database
User question
Embed
Retrieve top-k
Rerank
LLM + retrieved context
Answer with citations

Your documents are read once, split into passages, and stored so they can be found by meaning rather than by keyword. When somebody asks a question, only the passages that actually relate to it are pulled out and handed to the model alongside the question. The model answers from that text, and points back at the document it came from.

What we build

From a question to a working system

The work usually looks like one of these. All of it works in Khmer and English.

  • Document Q&A

    Ask one question across your contracts, invoices and policies, and get an answer that points at the page it came from.

  • Khmer OCR & search

    Turn photographed and scanned Khmer paperwork into text you can actually search through.

  • Chat assistants

    Assistants for your website or Telegram that answer from your real catalogue, and hand over to a person when they should.

  • Workflow automation

    The repetitive internal job — sorting, extracting, drafting, routing — done end to end, with a person approving anything that matters.

  • Task-specific models

    A small model fine-tuned for one narrow job, for when a general-purpose model is more than you need.

  • Private LLM deployment

    Open-weight models running on hardware you control, so your data never reaches anyone else's servers.

Where it runs

You choose where your data lives

The same promise as everything else we build: at the end it is yours. Source code, model weights, configuration and documentation, handed over in writing.

  • 01

    Cloud API

    The quickest and cheapest way to start. Prompts and documents are sent to a model provider, under terms we go through with you before anything is connected.

    Your data
    Leaves your network, under a provider agreement
  • 02

    Private cloud

    Open-weight models on a server rented in your name and configured by us. Heavier to run, and nothing is shared with a model vendor.

    Your data
    Stays inside infrastructure you own
  • 03

    Fully on-premise

    A machine in your own office. No internet connection is needed to answer a question. The most work to set up, and the only option that is airtight.

    Your data
    Never leaves the building

Tell us what you are building.

Bring the problem and whatever data you already have. If AI is the wrong tool for it, we will tell you that instead of selling it to you.

Get in touch