RUUZEN
AI

How to Build an AI Chatbot Using Free Cloud Service Plans

AI engineer Kartigan Vijay described the set of tools he used to create the sports chatbot Rune Score with minimal startup costs. The project combines development tools, a database, internet search, monitoring, and voice synthesis. However, the ability to use such a solution for free depends on the workload and the terms of individual services.

20 Nov, 14:42Read in Russian
entrepreneur.com

In an op-ed for Entrepreneur, the developer explains that behind the chatbot's simple interface lie several interconnected systems. They must retrieve up-to-date sports data, understand differently phrased questions, save conversation history, and handle errors when interacting with language models.

For project design, the author used Figma, and the code was hosted on GitHub. The repository served as a shared workspace for two AI tools with distinct roles: Claude Code handled development, while Codex reviewed changes and helped formulate subsequent tasks.

All improvements were made in separate branches. Before merging with the main project version, the author personally reviewed the changes and approved pull requests. This approach allowed maintaining control over the AI-generated code.

However, the described free access to Claude Code was temporary: the author used a guest invitation from a Claude Max subscriber. According to him, it provided a week of Pro-level access. Therefore, this component cannot be considered a permanent free resource for development.

Vijay deployed the chatbot's backend and interface on Railway, setting up a separate testing environment. In the article, he refers to startup credits for new users. Railway's current documentation distinguishes between a trial period with a one-time $5 credit for 30 days and a free plan with a monthly resource limit of $1.

On Railway's permanent free plan, services go to sleep during inactivity, and after exhausting the limit, they stop until the next billing period or a switch to a paid plan. This restricts its use for continuous operation of a growing project.

The author implemented user registration and login via Clerk. He stored conversations and application data in Neon—a cloud-based PostgreSQL database capable of reducing computational resource consumption during inactivity.

Langfuse was used to monitor the AI agents. It helped track model requests, failures, and processing costs. Information search was distributed between Exa and Firecrawl: the first service located pages, while the second converted their content into text for further analysis.

Before selecting a language model, the developer tested various options through Nvidia's API catalog, using credits provided for experimentation. However, the article does not disclose the full payment scheme for the model serving the finished chatbot's users. Thus, free testing does not confirm the absence of costs for subsequent operation.

Azure AI Speech provided voice responses. The author purchased the domain through Cloudflare, where he also configured DNS and security features. Registering a custom domain remains a paid service, although Cloudflare claims no markup over the registry's cost.

The described experience demonstrates how combining free quotas and trial offers can reduce the costs of creating and validating an idea. The future budget depends on the number of users, model requests, volume of stored data, and the need for continuous service availability.