AI drafts in pencil. You approve in ink.

• AI agents

  Agents that read a request, look things up, use your tools and APIs, and pause for approval before anything that can’t be undone. Our own engineering runs on coding agents every day.

  Running in: LLM Proxy

• AI workflows

  Multi-step automations across forms, CRM, payments, email and documents, with a person approving the steps that matter.

  Running in: Image Workflow

• LLM gateway and cost control

  One endpoint for many models and accounts, with pooling, usage caps, token refresh and a clear error when limits are reached.

  Running in: LLM Proxy

• AI grading and feedback

  Rubric scores, error labels and feedback that helps learners improve without handing them the answer.

  Running in: AutoGrader, IELTS Platform

• Copilots inside your product

  Inline suggestions, completions and grammar checks in the editor or form your users already work in.

  Running in: LaTeX Tool

• Semantic search

  Find videos, images and documents by what they mean, not only by the words they contain.

  Running in: MiTube

• Vision and OCR

  Turn scans, photos and video frames into structured, searchable data, with quality scores a reviewer can act on.

  Running in: Archive OCR, MiTube

• Private and on-premise AI

  Open-weight models on your own hardware, with encrypted storage and servers you control, for data that can’t leave the building.

  Running in: AutoGrader, DPFileSystem, DPInfrastructures

Work still runs on paper, spreadsheets and chat groups.

Digital transformation in steps your team can absorb: we map how work flows today, move it into software people will actually use, and train everyone as we go.

Customer details are scattered across people and tools.

A CRM set up around how you sell: leads from your website and Facebook in one place, follow-ups that send themselves, and AI that summarises each customer and suggests the next step.

Customers ask for advice at all hours, and staff can’t keep up.

An advisory chatbot, built as an AI agent, on your website, Messenger and Zalo: it answers from your own product information, recommends the right option, takes contact details and hands over to staff when needed.

Leads come in, then go cold.

An automated funnel: form, CRM record, payment request, reminders and confirmation, with nobody copying data between tools.

Grading and reviews take up your experts’ evenings.

AI drafts marks and comments against your rubric, so experts confirm instead of starting from a blank page.

Your product needs AI features people will actually use.

Copilots, semantic search and recommendations built into the screens your users already know.

Your data can’t leave your servers.

Private AI: open-weight models, a model gateway with spending caps, and encrypted storage on your own infrastructure.

Your rescue team still dispatches by phone.

A rescue platform under your own brand: customers send their location and see the price first, the nearest crew takes the job, and a dispatch board follows every truck until payment.

You need the software, not just the AI.

Web, mobile and desktop apps, with the backend and servers to run them. We build the whole thing.

Recent work for businesses

1. 1

   Find the right problem

   We sit with the people who do the work, map the process, and pick the step where AI saves the most time at the lowest risk.

2. 2

   Prototype on your data

   A working version on real examples, so you judge results instead of slides.

3. 3

   Build it to last

   Approval steps, logs, evaluation sets, cost caps and monitoring, so it holds up in daily use.

4. 4

   Run it and improve it

   We track quality and cost, update prompts and models as your data changes, and hand over everything you need to run it yourselves.