Case studies

What changed in the companies we worked with

AI SKILLS has trained 4,000+ people, worked with 200+ companies and organizations, and built 25 AI agents for businesses in Moldova and Romania. Below are real projects from 2025–2026: the problem, what we did, how long it took, and what changed. Where a client asked for confidentiality, we do not name them.

4,000+
People trained
200+
Companies and organizations
25
AI agents built
3
Languages: RO, RU, EN
Real estate

OneKey: an AI-powered CRM and website for a real estate agency

Client: OneKey, a real estate agency in Chisinau.

The problem
Client requests were getting lost between spreadsheets, messages and an off-the-shelf management system that did not match how agents actually work. Contracts, property sheets and offers were filled in by hand.
What we did
A CRM built exactly around the agency's processes, with AI: documents are generated from data automatically, scanned paperwork is read by the system, agents have a conversational assistant, and every client has an owner, a stage and a next step, with reminders and WhatsApp Business and telephony integration. We also built the agency's website.
Tools
React, Node.js, PostgreSQL, GPT-5 and Gemini 2.5, OCR, WhatsApp Business API
Timeline
4 months (July–October 2026)
The result
No client lost in spreadsheets or traditional management systems. The system was built precisely for their needs.
NGO · justice and anti-corruption

AJAM: the Digital Mirror, Moldova's first AI-powered digital mirror

Client: AJAM, the Association of Administrative Lawyers of Moldova, an NGO working on justice and anti-corruption, within the AI4JUSTICE / Face Your Future project supported by IDLO (International Development Law Organization).

The problem
AJAM wanted to talk to young people about integrity and the consequences of corruption in a different way than seminars: through an experience you live, not one you listen to.
What we did
The Digital Mirror: an AI kiosk designed and built specifically for this project, hardware and software, in which visitors hold an immersive conversation with an AI about integrity and ethical dilemmas. It is the first digital mirror in Moldova that runs on AI.
Tools
Custom-built digital kiosk, large language models, conversational interface
Timeline
2026
The result
Installed in innovation and university centers, where visitors talk directly with the AI about integrity.
Group of companies · contracts and tenders

Three AI agents for internal documents, international contracts and tenders

Client: A group of companies in Moldova with international contracts and frequent tender participation. Confidential, under a non-disclosure agreement.

The problem
Internal information was hard to find, deadlines in international contracts were tracked by hand, and preparing a tender application took days of reading specifications.
What we did
Three AI agents: the first answers questions about internal documentation with citations; the second monitors contracts, extracts deadlines, obligations, milestones and penalties, and sends alerts before due dates; the third analyzes the tender specification, builds the compliance matrix and proposes the structure of the response. A person reviews every step.
Tools
Claude and GPT-5, LangGraph, pgvector, n8n, Google Drive and SharePoint, EU hosting
Timeline
4 months (June–September 2026)
The result
Contract deadlines no longer depend on anyone's memory, and a tender application starts from a ready-made compliance matrix.
Distribution · car parts

A chatbot that finds the right part, in Romanian and Russian

Client: A car parts distributor in Moldova. Confidential.

The problem
Customers and staff lost time finding the right part for a make, model, year and engine, with the same questions repeated in Romanian and Russian.
What we did
A chatbot on the website and on Telegram that searches the parts catalog, answers in Romanian and Russian, and hands unclear cases to a human operator with the full conversation context.
Tools
LlamaIndex, Qdrant, GPT-5 and Claude, FastAPI, catalog integration via API
Timeline
5 months (May–September 2026)
The result
Customers get their answer in the chat right away, without waiting for an operator; operators handle only the ambiguous cases.
Retail, services, distribution

24/7 customer support agents on the website, WhatsApp, Messenger and Telegram

Client: Companies in retail, services and distribution in Moldova. Confidential.

The problem
The same questions came in through five channels, and in the evenings and on weekends response times stretched to hours.
What we did
Text agents that answer around the clock from the company's knowledge base in Romanian, Russian and English, check order status and escalate to an operator with the full conversation context. Customer feedback is collected in a structured way.
Tools
n8n and Make, OpenAI and Anthropic, RAG knowledge bases, CRM, helpdesk and e-commerce integration
Timeline
In production since 2025
The result
An instant answer at any hour, on every channel, with operators focused on the cases that really need them.
Public institutions and international organizations

AI training for institutions that work with sensitive data

Client: UNDP Moldova, the Superior Council of Magistracy, the Ministry of Justice, the Ministry of Internal Affairs, the Border Police, the Public Services Agency, CNAS, CTIF and others.

The problem
Staff were already using ChatGPT, with no rules for the personal data and sensitive information they handle every day.
What we did
Hands-on programs on responsible AI use: which data never goes into public tools, how to verify answers, how to automate repetitive tasks. For UNDP, a two-day training in Tiraspol for civil society organizations from the left bank of the Dniester (August 2026), followed by mentoring.
Tools
ChatGPT, Claude, Copilot, Gemini; materials in Romanian, Russian and English
Timeline
2025–2026
The result
Institutions with clear rules for AI use and staff who know what they can and cannot delegate to an AI assistant.

Frequently asked questions

Can I talk to a client who has worked with AI SKILLS?
Yes. References are provided on request, with the client's prior consent, usually before the contract is signed. Write to us at denis@aiskills.md.
Why aren't all client names shown?
Some projects are under non-disclosure agreements or involve sensitive data (contracts, medical data, customer data). We publish a name only where the client has agreed.
How quickly do results show?
A chatbot or a simple agent goes into production in a few weeks; a CRM or a multi-agent system takes 3–5 months. Training shows the next day: people leave with prompts and workflows they use immediately.

Want a similar result in your company?

The first consultation is free. In 30 minutes we tell you what can be automated and how long it would take.

Book the free consultation