Andrey Koptelov Showed How to Deploy an AI Agent in 8 Hours: From Prototype to Working System Without Budget

Imagine the situation: you've been tasked with "implementing artificial intelligence," but the budget is zero, data must remain within the company, and the results were needed yesterday. Meanwhile, the team has no programming experience and is only familiar with large language models through chat. Andrey Koptelov, an expert in business processes and corporate architecture, demonstrated how, under such conditions, to get a working AI agent prototype in just 8 hours, using only open-source software.

AI implementation often seems like a complex, expensive, and lengthy project, accessible only to large companies with unlimited resources. This deters small and medium-sized businesses and hinders initiatives within large corporations. But the reality is that modern tools allow for rapid and effective creation of functional AI prototypes without significant investment, proving the technology's value in practice.

The Reality of the Problem: AI is Needed, But Resources Are Not

The task of implementing AI is often set from above, but without a clear understanding of how to do it, especially when it comes to pilot projects. The main barriers are: lack of budget for expensive corporate solutions, the need to keep confidential data within the corporate perimeter (which excludes most cloud services), and the team's lack of programming skills or deep understanding of LLM operations. In such circumstances, the traditional approach to AI implementation becomes impossible, and the need for quick results only intensifies.

The Path to an AI Agent: Why a DIY Approach?

In conditions where there is no budget for ready-made solutions and deadlines are tight, the only way out is a "do-it-yourself" (DIY) approach using open-source software. The advantages of this path are obvious: zero licensing costs, full control over data (as everything runs locally), and the ability to quickly adapt and scale the prototype to specific needs. This allows not just "playing" with AI, but creating a real tool integrated into business processes, which can be refined and expanded as resources and understanding of tasks emerge.

How the AI Agent Was Designed: 7 Steps to a Prototype

Andrey Koptelov developed a step-by-step guide that allows for the creation of a functional AI agent prototype in a single workday (8 hours). The main idea is to use a combination of locally deployed tools to ensure autonomy and data security. The agent was designed to analyze and generate text, search for information, and integrate into simple automated processes, all without requiring programming.

The process consisted of seven clear steps:

1. Deploying a Local LLM in Ollama. Installing Ollama allows running large language models (e.g., Qwen) directly on a computer, without sending data to the cloud. This is critical for maintaining confidentiality. For less powerful PCs, models with :4b or :9b tags are recommended.
2. Deploying n8n Locally with Docker Desktop. n8n is an orchestrator that allows creating automated workflows without code. Docker Desktop simplifies its installation and launch in an isolated container.
3. Integrating n8n and Ollama. n8n has special nodes for interacting with Ollama. This allows connecting the local LLM with automated processes, such as parsing websites or analyzing text.
4. Writing Prompts. Clear and structured prompts (Role + Context + Task + Format + Example) ensure correct LLM operation. This step includes testing and iterative improvement of prompts until 80% accuracy is achieved.
5. Connecting RAG (Retrieval-Augmented Generation). If the agent needs to answer questions based on an internal knowledge base, RAG is used. This allows the model to refer to a "cheat sheet" (vector database) to obtain relevant and accurate information, avoiding "hallucinations" and loss of context in large documents.
6. Uploading Documents to Storage. After configuring RAG, documents are uploaded to the vector database, and the AI agent is ready to answer questions based on them.
7. Demonstrating the Prototype to the Manager. The final step is presenting the working prototype, which demonstrates the value of AI in a specific business process.

Limitations of this approach included: unsuitability for image/video generation (requires specialized software and powerful hardware), low performance under high loads (local LLMs are slower than cloud-based ones), and response delays (5-50 seconds).

Implementation and Results: A Working Prototype in One Day

The entire process, from deploying Ollama to automating a simple process, such as regular analysis of websites with subsequent information dissemination to stakeholders, was implemented in 8 working hours. If required, an internal knowledge base was also set up using RAG, allowing for a working assistant prototype that answers questions based on corporate documents without external data transfer.

The main result is the creation of a functional prototype that proves the concept and demonstrates the capabilities of AI without significant investment. This allows moving from theoretical discussions to practical application, showing the real value of AI for business.

How to Implement This in Your Company: A Quick Start Checklist

If you want to replicate this case and quickly launch a pilot project with an AI agent, here are the key steps:

• Define a process for pilot implementation. Choose a routine task related to text processing, analysis, or generation where AI can quickly show results.
• Deploy Ollama with a local LLM. Ensure your PC meets the minimum requirements (32 GB RAM, 4+ cores, 20+ GB free space) and choose an appropriate model.
• Install n8n for automation. Use Docker Desktop for quick and seamless installation.
• Integrate LLM and orchestrator. Configure n8n to interact with Ollama using special nodes.
• Teach the agent to work with prompts. Develop and test prompts to ensure the AI agent consistently performs its tasks.
• Connect a knowledge base via RAG (if necessary). If the agent needs data from internal documents, set up RAG to access your corporate information.

Don't strive for perfection from the first try. The goal of a prototype is to prove the value of AI in a specific process, not to replace people. This approach allows for quickly testing hypotheses and getting feedback for further development.

If this case sounds like what's happening in your company, our manager can help: he'll analyze your business and niche for free and point out where an AI agent would bring a real result in your case. Message the manager