Custom software & applied AI · Kazakhstan
Software and applied AI that work for your business
We build web services and backend systems, integrations and automation, LLM and agent-pipeline solutions, computer vision, GIS and real-time systems. Turnkey — from discovery to support.
- Ruby on Rails
- Python
- Go
- TypeScript
- LLM · agents
- PostgreSQL
Services
The full development cycle — from idea to operation
We own the outcome end to end: analysis, architecture, code, tests, deployment and support.
Custom software development
Web services, SaaS, portals and user dashboards, backend systems and APIs — from MVP to high-load product.
Applied artificial intelligence and LLM agents
Agent pipelines, AI assistants, document data extraction and content generation — with quality and cost control.
Integrations and business-process automation
We connect accounting systems, payments, e-signatures and messengers — and remove manual data entry and duplicate work.
Data processing and computer vision
License plate recognition, OCR, parsing of estimates and BIM models, analytics and reporting.
Geographic information systems and real-time systems
24/7 monitoring and control of distributed sites, offline maps, triangulation and operator workstations.
Discovery, concepts and prototypes
We analyze the specification, design the architecture and show the solution in a prototype before development starts.
Deployment, DevOps and support
We take systems to production and keep them running: containers, CI/CD, monitoring, backups and staff training.
Code audit, security and modernization
We review existing systems, close vulnerabilities, upgrade the stack and speed up the application.
Approach
AI you can trust
The model reads and classifies — deterministic code computes the numbers and decisions, and an expert approves the result. That is why every number is reproducible and traceable to its source.
Data
Documents, scans, spreadsheets, BIM models, video streams, accounting systems.
The model reads
Extracts facts, classifies, turns requirements into strict schemas.
Code computes
Arithmetic, rules and verdicts live in a deterministic engine.
An expert approves
Ambiguous cases go to a specialist; decisions are recorded.
Result
A report, an API response or an action — traced back to the source document.
The model reads — code computes
We never trust arithmetic, rules or verdicts to the model. That makes results reproducible and verifiable.
Expert in the loop
Doubtful cases are never published automatically: a specialist reviews them and decisions are recorded.
Traceability
Every number leads to its source: a document, row, cell, model element or transaction.
Works without the cloud
CPU models, offline maps, closed networks — client data never leaves the client's infrastructure.
Quality by default
Automated tests, linters, security checks and CI are part of every project.
Predictable AI costs
Caching, token budgets and per-agent cost accounting — no surprises on the invoice.
Industries
We work across every area of digital transformation
Education, retail, finance, construction, industry, security, government and more.
Projects
Featured projects
Production systems, prototypes and products. Client names are withheld under confidentiality terms.
Digital platform for a learning center with an AI teaching assistant
Learning center, Kazakhstan
A website, a course catalog and four dashboards — student, parent, teacher and administrator — with AI generation of learning materials, live lessons and payments.
- 805automated tests
- 4role-based dashboards
- Ruby on Rails 8.1
- Claude API
- PostgreSQL
- Hotwire
HR platform with multi-agent artificial intelligence
HR Tech product, SaaS for companies
Automation of the entire employee lifecycle: 16 AI agents cover hiring, onboarding, performance reviews, learning, engagement and compliance.
- 1,005automated tests
- 16AI agents under an orchestrator
- Ruby on Rails 8.1
- RubyLLM
- React
- TypeScript
AI agent for financial covenant compliance checks
Banking AI challenge
The agent reads a corpus of credit documentation and a transaction ledger and, for every covenant of every borrower, returns a verdict, the actual value and the evidence transaction.
- 97.2%match with the reference: 35 of 36 cells
- 84cells in the hidden set with no critical findings
- Python
- Anthropic SDK
- Claude vision
- PyMuPDF
Command center for a network of counter-UAV systems
Developer of site protection systems
Software for centralized monitoring and control of a network of stationary drone detection and jamming systems from a single command center — offline, 24/7.
- 70–6000 MHzdetection and direction-finding range
- 24/7operation without internet
- Go
- WebSocket
- SQLite
- MapLibre GL
Night-time license plate recognition on ordinary IP cameras
Operator of a parking SaaS platform
The module raises night-time recognition accuracy for Kazakhstan license plates from ~87% to 97.5% — without dedicated ANPR cameras and without the cloud.
- 97.5%night accuracy vs 86.5% for the baseline
- 0%false reads
- Python
- PyTorch
- ONNX Runtime
- OpenCV
Linking completed quantities to the BIM model via estimates and certificates
Major construction group
The prototype links lines of cost estimates and completion certificates to specific BIM elements and tracks designed, claimed and accepted quantities with full traceability.
- 99.4%link accuracy on the independent test (target ≥ 95%)
- 97.2%lines linked automatically (target ≥ 70%)
- Python
- IfcOpenShell
- Claude API
- three.js
How we work
Turnkey delivery in six clear stages
Every stage ends with a measurable result you accept before we move on.
-
01
Discovery
We study processes, documents and constraints, and define goals and success criteria.
-
02
Design
Architecture, scenarios, a UI prototype, a plan and an estimate — you see the solution before development starts.
-
03
Development
Short iterations with demos and a staging environment from the first weeks.
-
04
Testing
Automated tests, AI control sets, security checks and acceptance testing.
-
05
Deployment
On your servers, in the cloud or in a closed network — with monitoring and backups.
-
06
Support
SLA-based support, product evolution, documentation and staff training.
Technology
A production-proven stack
We pick technology to fit the problem, not the other way round. Our core is Ruby on Rails, Python and Go.
Languages and frameworks
- Ruby on Rails 8
- Python
- Go
- Ruby
- TypeScript
- JavaScript
- Node.js
- FastAPI
- Pydantic
- Uvicorn
- Jinja2
- Jekyll
LLMs and agents
- Claude API
- Anthropic SDK
- Claude Agent SDK
- Claude Code
- MCP
- Tool use
- Structured outputs
- Prompt caching
- Batches API
- RubyLLM
- DeepSeek
- Kimi
ML and computer vision
- PyTorch
- ONNX
- ONNX Runtime
- NumPy
- OpenCV
- YOLO
- CRNN · CTC
- Pillow
- Tesseract.js
- pgvector
Rails ecosystem
- Hotwire
- Turbo
- Stimulus
- Importmap
- Propshaft
- Vite Ruby
- Phlex
- ViewComponent
- Devise
- OmniAuth
- Authentication Zero
- Rodauth
- JWT
- Pundit
- CanCanCan
- Action Policy
- AASM
- Pagy
- Kaminari
- Ransack
- Jbuilder
- money-rails
- Draper
- Interactor
- dry-validation
- Rack::Attack
Frontend and mobile
- React
- Inertia.js
- Vite
- Tailwind CSS
- Headless UI
- Framer Motion
- Recharts
- Chartkick
- React Hook Form
- Zod
- htmx
- CodeMirror
- PWA
- Service Worker
- Web Push
Databases and search
- PostgreSQL
- SQLite
- JSONB
- pg_search
- pgvector
- Meilisearch
- Redis
- Active Storage
- S3
- Cloudflare R2
- MinIO
Queues and real time
- Solid Queue
- Solid Cache
- Solid Cable
- Action Cable
- Turbo Streams
- WebSocket
- Sidekiq
- GoodJob
- Mission Control
Documents and reports
- PyMuPDF
- pdf-reader
- python-docx
- docx
- openpyxl
- Roo
- caxlsx
- Excelize
- Prawn
- go-pdf
- Redcarpet
- CSV
GIS, BIM and 3D
- MapLibre GL
- PMTiles
- MBTiles
- three.js
- IfcOpenShell
- IFC4
- BCF 2.1
- Cytoscape.js
Integrations and payments
- Kaspi
- Halyk Epay
- Freedom Pay
- CloudPayments
- 1С
- PKI / X.509
- Google OAuth
- Sign in with Apple
- Google Calendar API
- Slack
- Microsoft Teams
- Docuseal
- Resend
- Faraday
Testing and quality
- RSpec
- Minitest
- FactoryBot
- Capybara
- Selenium
- Playwright
- pytest
- Vitest
- WebMock
- VCR
- SimpleCov
- RuboCop
- Brakeman
- bundler-audit
- Bullet
- ESLint
- ShellCheck
- bats
DevOps and infrastructure
- Docker
- Kamal 2
- Thruster
- Puma
- GitHub Actions
- GitLab CI
- Azure DevOps
- Hetzner
- Cloudflare
- systemd
- Windows Service
- Sentry
- Lograge
- rack-mini-profiler
- GitHub Pages
Have a challenge? Let's talk
Describe the process you want to automate or the product you need to build. We reply within one business day with questions and a proposed engagement format.