Jev AI Use Cases: 3 Projects from the AdaL Hackathon

What can you build with Jev?

At AdaL's “What Would Jev Do?” online hackathon, held October 6–11, 2026, developers explored that question through working applications. Glasser supported the event as a partner, including a dedicated Best Use of Glasser award. This article highlights three participant projects, rather than announcing award winners. About the hackathon →

The projects address different problems, but each gives Jev a focused role: interpret an input and return a structured judgment that software can use. Code, external data, and other models handle the surrounding work.

Here is what those Jev AI use cases look like in practice.

1. README Clew 2: Classify Documentation Claims Before Checking Them

A repository evolves. Its README does not always keep up.

Earl Grey built README Clew 2 to check documentation against public JavaScript and TypeScript repositories. The application traces supported README claims back to code evidence and produces a claim-by-claim report.

Jev's role is classification, not verification. Accessed through Glasser, it identifies whether a line concerns a dependency, command, environment variable, file, or another category. Five deterministic verifiers then check supported claims against repository files. Claude Haiku summarizes the findings.

The report distinguishes agreement, contradictions, omissions, and claims the tool cannot verify. Uncertain classifications can remain unresolved instead of becoming accusations that correct documentation is wrong.

For builders, the useful pattern is the separation of responsibilities: let a model interpret the language, then use code to check facts that have an exact answer. The output makes that distinction visible through line-level evidence.

Read Earl Grey's build story and try README Clew 2 →

2. Scam Listing Detector: Turn Rental Warning Signs Into Inspectable Signals

A rental description rarely contains everything someone needs to assess it.

After encountering suspicious listings during an apartment search, Sandeep Manoharan built Scam Listing Detector, a Chrome extension that adds risk badges as users browse. Hovering over a badge reveals the reasons behind the assessment.

Jev answers eleven focused yes-or-no questions and returns probabilities. The application applies its own weighting rather than asking the model for an unrestricted verdict.

Glasser supplies supporting information: web scam reports, address lookups, reverse image searches, and public property records. The builder also accesses Jev through Glasser.

The result brings model judgments and external evidence into the browsing experience. Users can see which listings deserve closer investigation and why.

The boundary matters: a risk signal is not proof of fraud. As the builder emphasizes, the extension supports further checking—not accusations against a person or listing.

This project illustrates a second pattern: focused model questions become more useful when the application can retrieve relevant evidence and explain its warning.

Watch Sandeep's demonstration → · Explore the source code →

3. Helios: Organize Competitor Ads Into Comparable Messaging Categories

Reading one competitor ad is straightforward. Following changes across many ads takes more structure.

Helios, shared by GitHub user shrutidoshi94, retrieves Meta and LinkedIn ad-library data through Glasser. It organizes ads by theme, audience, and claim verifiability, then compares weekly snapshots.

Jev supplies structured classifications, accessed separately through OpenRouter. Code handles the comparisons and brief generation. Low-confidence tags can be held for review.

The dashboard lets users inspect ads, compare messaging patterns, and download a brief. Data responses are cached, live calls require an opt-in, and a configurable budget cap limits spending.

The hosted demo uses saved responses and keyword-based classification; local configuration is required for live Glasser retrieval and Jev. Historical coverage is limited, so its comparisons are not a complete record of a competitor's advertising.

For builders, the pattern is classification followed by reproducible analysis: a model interprets the message, while code measures changes within the collected dataset.

Explore Helios, its setup instructions, and hosted demo →

What These Jev Use Cases Have in Common

These are not three versions of the same chatbot. They place a focused judgment inside a larger application:

The distinction is useful when choosing a project. Start by identifying the judgment that requires interpretation. Then decide which parts need external information, exact rules, calculations, or a written explanation.

For background on the model, read Jev AI Explained: How It Works and When to Use It.

Start With One Question Worth Answering

A useful Jev project begins with a specific question—not a requirement to put AI into every step.

Choose an input, define the possible answers, and decide what the application should do when the judgment is uncertain. Keep the evidence available so users can inspect the result.

These hackathon projects offer three starting points: checking documentation, investigating a listing, and following advertising messages. Glasser contributes model access or external data where the workflow needs it; the builders decide how those inputs become something useful.