I'm a 25-year-old AI consultant and I made $50,000 in 2 months. Here's my advice for companies trying to adopt AI.

I've made over $50,000 since December building AI tools companies can use to grow. My main challenge: managing expectations about what AI can do.

• Sasha Aptlin has earned more than $50,000 building custom AI software for companies. 
• The 25-year-old's mission is to help clients make more money using large language models. 
• Here's what it's like to be an AI consultant. 

This as-told-to essay is based on a conversation with Sasha Aptlin, a 25-year-old AI consultant based in Dubai and New York, about how he helps companies use large language models, or LLMs, to grow. Business Insider verified his earnings and client work by reviewing PDFs of his contracts. This story has been edited for length and clarity.

I started my machine learning career working on projects related to computer vision, natural language processing, and recommendation systems. Now, I run a consultancy that tries to help companies grow using AI — and business has been booming.

My entrepreneurial journey began in January 2022 when I founded an AI consultancy services firm called Aptford. During the first few months, I worked on small contracts where I helped clients understand how to use natural language processing — a type of programming that enables computers to understand, interpret, and generate human language — and built proof of concepts for AI systems.

But, demand for my services jumped when ChatGPT came out in November 2022. Founders and entrepreneurs saw the potential in using generative AI to save time and boost profits. They then came to me because they didn't have the technical skills to implement that effectively.

Initially, most people were interested in building customer support agents for their websites, which weren't big projects. Now, clients want to go beyond chatbots to create custom-made AI software that can grow their revenue.

For instance, I built an AI assistant for a sales training company that sends skills assessments to sales reps, processes them automatically, and then uses those insights to provide personalized improvement plans for each employee and team. I also developed an AI coach for a startup accelerator that offers advice to founders and a transport management system for a freight broker that automates their workflows.

One of the most exciting projects I worked on was for a midsize mortgage lender based in San Francisco. I built an internal AI platform that we fine-tuned with legal documents and other company data which loan officers could use to generate answers to customer support questions in real time. It cut response times from 20 minutes to just a few.

It's clear that companies are willing to pay for custom AI. From late December 2023 through to early February, I made nearly $50,000 from two contracts. I expect to receive another $12,000 from a client who postponed its contract and additional income through consultations.

Each project typically has three phases. In the first, I look through the client's workflow and create an action plan that defines the project's scope and identifies its risks. After that, my team and I build the software with its key features and tweak it based on the client's needs. Once that's done, my team and I reflect on our work and explore new ideas that can be used for the next project.

Most projects use AI models from leading LLM providers like OpenAI, Anthropic, and Cohere.

On average, I can complete a project in around 4 weeks, though it can take up to six weeks, depending on the complexity of the system I'm building.

The main challenge I face is expectation management. Clients have wanted me to complete projects that typically require weeks in a day, and some weren't satisfied with the final product.

There have been a few cases where the requests were simply unfeasible. Once, someone asked me for a system similar to the AI companion in Spike Jonze's movie "Her." Another time, a client wanted customer support agents that could replace its human workforce.

Clients have also withdrawn from projects because of concerns about data scraping and legal compliance.

My main advice for businesses interested in adopting AI is to familiarize yourself with large language models and how they could be used to address specific pain points. After that, collect as much data as possible — from meeting recordings to marketing materials — that can be used to fine-tune the AI models. From there, start experimenting with ready-made services like chatbot builders.

Getting your hands dirty will help you see the possibilities of what the technology can do.