Auto video

• 

Inspiration

Most online tests give users an interesting label, but rarely help them make a real decision. We wanted to build something more useful: an AI-powered consultation service that first understands the user’s situation, asks targeted follow-up questions, and then produces a personalized conclusion with clear reasoning and actionable next steps.

At the same time, we found that long reports are difficult to read and even harder to share. This inspired us to turn each personalized report into a short, structured video automatically.

What it does

Auto Video is an AI-powered assessment and personalized content generation platform.

Administrators can create different assessment scenarios, such as career direction, suitable majors, side-hustle opportunities, project viability, or personal strengths. Each scenario is driven by a configurable JSON workflow that defines questions, options, conditional branches, AI follow-up rules, prompts, and report structures.

After a user completes the assessment, the system:

1. Analyzes the user’s answers and identifies missing information.
2. Asks additional questions when necessary.
3. Generates a personalized assessment report.
4. Produces clear conclusions, confidence levels, risks, and recommended actions.
5. Converts the report data into a structured short video that can be viewed or shared.

Instead of generating a generic result, the system aims to help users answer practical questions such as: “Should I continue this project?”, “Which direction fits me better?”, and “What should I do during the next 30 days?”

How we built it

We built the platform with Next.js and designed the assessment engine around versioned JSON configurations.

Each configuration contains:

• Multi-step questions and answer components
• Conditional routing between questions
• Required information fields
• AI follow-up strategies
• Prompt templates
• Report sections and chart definitions
• Video scene and content mappings

The AI layer is responsible for understanding user responses, detecting missing decision-critical information, generating follow-up questions, synthesizing conclusions, and producing structured report data.

For video generation, we use a data-driven scene system. The AI output is converted into predefined scenes such as covers, lists, comparisons, charts, conclusions, and action plans. These scenes are rendered into videos through Remotion.

This architecture separates content, logic, visual presentation, and AI prompts, allowing new assessment scenarios and video styles to be created without rebuilding the entire application.

Challenges we ran into

One of the biggest challenges was preventing the assessment from becoming a generic AI questionnaire. Early versions included repetitive questions, unnecessary follow-ups, and conclusions that sounded reasonable but did not provide enough decision value.

We also found that users often provide incomplete or ambiguous answers. The system therefore needed to determine which information was essential before producing a conclusion, rather than asking follow-up questions without a clear purpose.

Another challenge was keeping the short report, full report, charts, and video consistent. Since all of them are generated from the same user data, conflicting conclusions can quickly reduce trust.

Automatic video generation introduced additional problems. Purely AI-generated layouts were visually inconsistent, while rigid templates felt repetitive. We had to balance structured scenes with enough variation to make each video feel personalized.

Accomplishments that we're proud of

We created a reusable assessment engine instead of building every test as an independent page.

The platform supports configurable workflows, conditional branches, version management, AI clarification, personalized reports, charts, and automatic video generation within one system.

We are especially proud that the system focuses on decision support rather than entertainment-only labels. Reports can include a direct recommendation, supporting evidence, uncertainty, potential risks, and a practical action plan.

We also built a structured bridge between AI-generated content and programmatic video rendering. This makes it possible to transform one assessment result into multiple delivery formats without manually editing every video.

What we learned

We learned that a useful AI assessment is not defined by how many questions it asks, but by whether it collects enough information to support a decision.

AI should not ask follow-up questions simply because it can. Every question should reduce uncertainty or change the final recommendation.

We also learned that personalization does not require completely unrestricted generation. A well-designed structure, combined with controlled AI output, often produces more reliable results than asking the model to generate everything freely.

For video generation, templates are not the enemy. The real problem is using templates without enough content-aware variation. A strong system needs stable visual rules, multiple scene types, and dynamic pacing based on the report content.

Most importantly, users value clear judgments and specific actions more than long explanations. A shorter report with a clear recommendation can be more useful than a detailed but indecisive analysis.

What's next for Auto Video

Next, we plan to improve both the assessment engine and the video generation system.

We will introduce stronger decision-quality checks so the system can evaluate whether it has enough evidence before generating a conclusion. Reports will include clearer confidence ranges, competing explanations, risk indicators, and measurable action plans.

For video generation, we plan to add more visual themes, adaptive scene selection, smoother transitions, richer charts, voice-over generation, subtitles, background music, and platform-specific export formats.

We also want to support follow-up assessments, allowing users to return after 7 or 30 days, submit new progress data, and receive an updated report and comparison video.

Our long-term goal is to make Auto Video a configurable AI consultation and content-delivery engine: one system that can understand a user, support a real decision, generate a personalized report, and automatically turn the result into clear, engaging visual content.

Built With

• front