I Built a Real Data App with Claude Code (and It Cost $0)

From Demo Illusion to Production Reality

Most demonstrations of AI-assisted coding showcase polished interfaces backed by simulated data, leaving aside the friction that appears when real databases, authentication, and deployment enter the picture. This video takes the opposite path. Tech With Tim sets out to build a functional application powered by genuine data, hosted at a live URL, and operated without any recurring infrastructure cost. The premise is deliberately constrained: no database bill, no hosting bill, and no AI credits spent during the development process. That constraint shapes every decision in the project, from the choice of data platform to the way Claude Code is used for planning, coding, and deployment.

The result is an exercise in applied AI engineering rather than a superficial showcase. The video walks through a complete workflow that starts with an idea, moves through prompt-driven planning and iterative implementation, connects to a hosted data source, integrates AI-powered analysis, and finishes with a publicly accessible application. Each stage surfaces practical considerations that only become visible when the data is real and the deployment target is not localhost.

The Central Constraint: Real Data, Zero Cost

The project begins with a deliberately strict budget. The goal is to avoid paid infrastructure while still working with a meaningful dataset and a live environment. This constraint excludes many common shortcuts. Instead of relying on a managed database with a free tier that quietly expires, the video uses Databricks free edition as the data foundation. The choice matters because Databricks provides a legitimate cloud workspace with enough functionality to store, query, and process data without requiring a credit card or generating surprise charges during development.

The zero-cost rule also extends to the AI layer. Claude Code is used heavily throughout the project, but the implementation is structured so that expensive API calls are not necessary. The workflow emphasizes careful planning before generating code, which reduces wasted tokens and prevents the kind of trial-and-error loops that quickly consume credits. By treating the AI assistant as a planning partner first and a code generator second, the video demonstrates a cost-aware approach to building with large language models.

Using Claude Code for Strategic Planning

The video dedicates significant time to the planning phase, which is a notable departure from the usual pattern of jumping straight into code generation. Claude Code is used in plan mode to map out the application architecture, identify the data flow, and decide which components are necessary before any implementation begins. This stage is presented not as a formality but as a critical step that determines whether the project remains within its zero-cost constraints.

The planning dialogue covers the overall structure of the app, the way data will be extracted from Databricks, the role of AI analysis, and the deployment strategy. By resolving these questions in advance, the developer avoids the expensive detours that characterize unplanned AI coding sessions. The plan mode output becomes a reference point for the rest of the build, and the video shows how each subsequent phase maps back to the decisions made during planning.

Connecting to Databricks Without a Credit Card

The data layer is built on Databricks free edition, which is used to host and manage the dataset that powers the application. The video shows the process of creating a workspace, setting up access, and preparing the data for querying. The emphasis is on using the platform's real capabilities rather than treating it as a mock backend. This means configuring credentials, understanding the query interface, and ensuring that the connection between the application and the data platform is stable enough for a live deployment.

The connection step introduces a practical challenge that is absent from localhost demos: the application must authenticate against an external service and retrieve data over the network. The video addresses this by walking through the necessary configuration and showing how Claude Code can assist with the setup. The result is a data pipeline that works in production, not just in a controlled local environment.

Integrating AI Skills Into the Data Flow

Once the connection to Databricks is established, the video moves on to the AI analysis layer. The goal is to add intelligence to the application by enabling it to process the real data and extract insights that a static front end could not provide. This is where the distinction between simulated AI features and genuine AI integration becomes clear. The application must handle actual query results, manage response formatting, and deal with the variability of real-world data.

The implementation focuses on making the AI skills practical and bounded. Instead of a loose natural-language interface that could produce unpredictable output, the analysis layer is designed to work within the structure defined during the planning phase. The video shows how prompts are crafted to produce useful output while remaining reliable enough for a public-facing application. This balance between flexibility and predictability is a central theme of the build.

Handling Data Refinement and AI Output

The middle section of the video deals with the transformation of raw data into something the application can use. Real data arrives with inconsistencies, missing values, and formatting quirks that simulated data never has. The video shows how these issues are handled within the context of the application, including cleaning steps, aggregation decisions, and the logic that converts raw query results into meaningful displays.

The AI analysis component is integrated into this refined data flow. Instead of operating on the raw dataset directly, the analysis queries the prepared data and generates observations that fit the application's purpose. The video demonstrates how to structure these AI calls to avoid runaway costs, emphasizing single-purpose prompts and bounded outputs. The result is an application that feels genuinely intelligent without depending on excessive API usage.

Building and Deploying a Live Application

The final stage moves from data analysis to the actual application. The video covers the construction of the front end, the integration of the data and AI layers, and the deployment to a publicly accessible URL. The deployment step is significant because it confirms that the zero-cost constraint has not been broken. The hosting solution is chosen specifically to avoid recurring charges, and the video shows how the application is published without signing up for a paid platform.

The live deployment also serves as validation for the entire workflow. The app is not a localhost demo or a set of static screenshots; it is a functioning service that pulls real data, applies AI analysis, and presents results through a usable interface. The video closes by reviewing the final product and reflecting on the trade-offs involved in building under a zero-cost constraint, making it a useful case study for developers who want to apply AI coding tools to projects with genuine production constraints.