Inspiration
Creators often reject a topic after seeing a crowded search page. But content count is not the same as solved demand.
A search for “filament dryer comparison” may return dozens of videos while comments still ask for power cost, humidity recovery time and multi-material tests. The broad topic was covered; the actual audience question was not.
Comment analysis tools tell creators what people discuss. Keyword tools tell them how much content exists. We built DemandRadar to expose the contradiction between those two signals.
What it does
DemandRadar imports audience comments and existing content supply, then:
- extracts explicit requests from multiple channels;
- groups requests into demand areas;
- maps the specific details audiences ask for;
- audits which details existing videos cover;
- flags false saturation when relevant videos exist but requested details remain unanswered;
- ranks the narrow angles worth creating.
Every result shows both sides of the proof: audience quotes and existing videos. It then highlights the unresolved details and creates a focused angle—not a generic topic suggestion.
Judges can run the bundled 3D-printing analysis with no account or key. Creators can also enter a YouTube Data API key and channel handles to analyze real recent videos, comments, supply results and source permalinks directly in the browser. CSV import and JSON evidence export provide a local workflow.
How we built it
DemandRadar is a zero-dependency, local-first browser application using semantic HTML, responsive CSS, JavaScript modules and the YouTube Data API v3.
The engine preserves source IDs, extracts requests, groups cross-channel demand, normalizes supply records and compares requested details with covered details. Code computes every metric:
- Demand: frequency, channel span, likes, explicit intent and recency;
- Apparent saturation: relevant supply, best match, freshness and result coverage;
- Detail gap: requested dimensions absent from all supplied content;
- Angle score: demand weighted by both topic whitespace and detail gap.
A topic is false-saturated when relevant supply is abundant but at least half of its requested details remain uncovered.
This boundary is deliberate: models may help read text in a future version, but quotes, links, identities and arithmetic stay controlled by code.
Challenges we ran into
Our first version was another evidence-backed idea recommender. It worked, but the central pieces—comment clustering, scoring and content briefs—already existed elsewhere.
We challenged the product premise instead of hiding that overlap. The useful unanswered question was not “what topic is popular?” It was “why does demand persist after apparently relevant content already exists?”
That led us to separate topic saturation from detail coverage. A topic can be crowded and still contain a high-value unanswered angle. Modeling that contradiction made the recommendation both more specific and more defensible.
The second challenge was trustworthy evidence. Audience quotations are selected by source ID rather than generated, and imported data remains entirely in the browser.
Accomplishments that we're proud of
- A narrow false-saturation detector rather than a generic idea generator;
- side-by-side proof that supply exists and audience demand persists;
- explicit missing-detail extraction;
- deterministic and inspectable scoring;
- separate comments and supply CSV imports;
- source identity preserved through UI and JSON export;
- a complete local workflow with no build, account, key or cloud dependency;
- a four-mission interactive tutorial that teaches the workflow in 30 seconds;
- 23 automated assertions covering live-mode validation, tutorial safety, automatic supply analysis, the evidence pipeline and custom domains.
What we learned
A topic is not answered merely because a video with matching keywords exists.
The most valuable content opportunities can hide inside crowded categories. The opportunity is often a missing constraint, measurement, comparison or real-world condition—not an entirely new subject.
We also learned that originality comes from asking a sharper question, not combining more AI features.
What's next
The production version will:
- collect relevant videos and comments through the YouTube Data API;
- transcribe existing videos;
- extract atomic audience requirements using schema-validated semantic models;
- verify each requirement against transcript passages;
- cite the requesting comment and closest supply timestamp side by side;
- monitor whether the same unmet detail persists over time.
Built With
- api
- css3
- deterministic-scoring
- github
- google-cloud
- html5
- javascript
- local-first-processing
- node.js
- render
- youtube
Log in or sign up for Devpost to join the conversation.