Applied AI in real workflows

The first question is not cloud or local. It is which process is worth it.

We start from one of your processes, measure it before and after, and deploy where the number justifies it. Software, machine and infrastructure come from one place.

For AI to work inside a company you need all three: the software that connects it, the machine it runs on, and the infrastructure underneath.

That is why it comes from one supplier here, not four. A software house will write the integration but will not put the machine in. A hardware supplier sells the server and stops there.

How a project starts

Most deployments fail not because the model is wrong, but because nobody said in advance what working means. So we start from the measurement, not the tool.

  1. 01

    We pick one process

    High volume, repetitive, with a knowable correct answer and a cheap error. And it must already have an archive of handled cases.

  2. 02

    We measure against your archive

    We take two hundred real cases you have already handled, run them, and count. The number arrives before you commit to anything.

  3. 03

    We deploy with a person in the loop

    Nothing leaves without your people approving it. The same check stays in place afterwards, on every change.

Services

AI & LLM Solutions

We integrate modern AI systems into business workflows, internal tools, and intelligent assistants for practical operational use.

  • Custom AI assistants and chatbots
  • LLM integration with business processes
  • RAG systems with private documents
  • Local AI deployment for privacy-sensitive operations

Software Development

We design and build practical, production-ready software systems for business operations, technical workflows, and custom integrations.

  • Web applications and business platforms
  • Backend services and APIs
  • Automation tools and system integrations
  • Databases and operational dashboards

IT Infrastructure

We design and deploy reliable infrastructure with focus on security, documentation, and long-term maintainability.

  • Proxmox VE virtualization and hypervisor management
  • OPNsense firewall and network security
  • Wazuh SIEM and security monitoring
  • Backup, disaster recovery, and system administration

Hardware Supply

We supply professional hardware and software for technical, business, and infrastructure deployments across Bulgaria and the EU.

  • Servers and virtualization hosts
  • Workstations and business computers
  • Network equipment and infrastructure
  • Backup and storage systems

Quantum Readiness Platform

Post-quantum cryptography readiness, built and operated by Elkan Electronic

QRP discovers and inventories cryptographic assets across networks, hosts, repositories and CI/CD, scores post-quantum risk with evidence references, and produces a migration plan. It is deployed on the organisation's own infrastructure.

  • Cryptographic inventory across network, hosts, repositories and CI/CD
  • Deterministic, explainable scoring with evidence references
  • Evidence for DORA Article 6
  • On-premises deployment, no external LLM, data never leaves the organisation

Free first DORA report

For EU financial entities under DORA's full ICT risk framework. Run our open-source scanner, send us one masked file you have read first, and within two working days get a PDF that maps the findings to Articles 6 and 7(4) of RTS 2024/1774. See a sample report before you ask, no form.

Get the free DORA report →

Baton — task logbooks for Claude Code

Work that survives the session. An open-source plugin for Claude Code, built by Elkan Electronic.

An AI agent forgets the work between sessions. Baton keeps each task in a folder with a logbook, and hooks that Claude Code runs, not the agent's goodwill, make sure the logbook is read at the start and written at the end.

  • One folder and one logbook per task, plain Markdown on your machine
  • Each session opens on the board: what is waiting, and on whose move
  • A turn that leaves work unrecorded is handed back to the agent
  • Runs where Claude Code runs hooks: terminal, IDE, the desktop Code tab
Baton: these task folders hold work newer than their LOGBOOK.md:
  - market-research

Prepend an entry to each one's LOGBOOK.md before finishing — what was asked, what was done, the result, and what is still open. …

→ agent writes:  ## 2026-10-27 16:40 — compared three scenarios

What it does not do: it cannot tell whether an entry is true, and in claude.ai chat its hooks do not run.

Set it up for your team: office@elkan-electronic.com