AI implementation for business in Kazakhstan
I help companies put AI into working processes: from the audit and task selection to a system running in production. I design the architecture, build the pilot, take it to production, and measure the effect.
I work with Kazakhstani companies on-site (Almaty) and online, and with everyone else online. The stack: LLMs, RAG, AI agents, local models, and integrations with your systems.
Who this is for
- Companies with a heavy flow of documents, requests, or tickets
- Product teams adding AI features to their product
- Executives who need an AI strategy and prioritization
- Organizations with strict data-privacy requirements, up to a fully local setup
Tasks I solve
- Document processing and classification
- Customer support and internal help desks: LLM-based assistants
- Search and answers over a company knowledge base
- Automation of routine process steps with AI agents
- Report preparation and data structuring
- Copilots for staff inside their working tools
What I implement
How implementation works
Five steps, each with a concrete deliverable.
- 01Audit
We map processes and data, and find automation candidates
- 02Impact estimate
We estimate the effect and cost, and pick the pilot process
- 03Pilot
I build a working prototype on one process with real data
- 04Production
Integration with your systems, access control, security, team onboarding
- 05Quality control
Evaluation, monitoring, support, and a development plan
Engagement formats
Each format ends with a concrete result.
| Format | Result |
|---|---|
| AI audit | A process map, automation candidates, an estimate of effect and risks |
| AI strategy | Architecture, roadmap, budget, priorities, and KPIs |
| Pilot | A working prototype on one process |
| Implementation | Integration with data, CRM, ERP, documents, and internal systems |
| AI governance | Policies, access control, evaluation, security, and quality control |
| Technical due diligence | An assessment of an AI product, team, architecture, and risks |
Data security
Your data does not have to leave for someone else's cloud. We choose the setup during the audit: cloud, hybrid, or fully local, based on your requirements.
- On-premise local models: data stays inside your infrastructure
- Privacy-by-design: data-flow mapping, PII and access control
- Hybrid schemes: sensitive data stays local, the rest goes to the cloud
- Finetuning and quantization for ordinary hardware, no GPU farm required
Proof instead of promises
Author of published AI work: knowledge transfer into small models (arXiv:2502.08213), the MPT-VC neural codec, open models on Hugging Face.
Levitan, on-device voice input, runs for real users on Windows and macOS. A product, not a demo.
I build AI strategy, teams, and governance inside an operating company, so I know how AI takes root in an organization.
Implementation questions
Can we implement AI without sending data to a cloud?
Yes. I select and finetune local models that run inside your infrastructure, down to CPU-only setups without a GPU cluster.
We have no ML team. Is that a problem?
No. I design the system so your current team can operate it, and train them when needed.
How long does a project take?
It depends on the process and the data. An audit and a pilot take weeks, not months. I give the exact estimate after the audit.
Do you work outside Kazakhstan?
Yes, online. On-site formats are available in Almaty.
Where do we start?
With a consultation: we discuss the task and decide whether an audit is needed. Write to libr@bk.ru.
Discuss implementation
Describe the process you want to automate: what the data is, who works with it, and what counts as a result. I will reply with what I can do.