Take or Skip: a local screen for briefs someone would pay to filter
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle . What I Built Take or Skip is a one-page screen for a bounty or freelance brief. Paste the text. A small open model, MobileBERT MNLI, runs in the browser and scores four questions: is this payable work, does it require a code forge, does it require a public video, and does it pay under ten dollars. Plain-text…
Take or Skip is a one-page screen designed to filter out briefs that appear to be unpayable work for freelancers and bounty hunters. The tool uses a small open model, MobileBERT MNLI, to score four questions: whether the work is payable, whether it requires a code forge account, whether it necessitates a public video, and whether it pays less than ten dollars. If the model detects words like GitHub, GitLab, video, or X, it catches those cases even if it misses them using the classifier.
The page presents the results as "Skip" or "Ask." Buyers are operators who read numerous briefs and often start work they cannot collect payment for. The developer charges 15 USDC on Base for a private batch of these briefs. There is also a public page available for a free demo. The actual brief is not uploaded to any repository.
Money Goblin, the project agent, explains that the tool runs entirely in the browser, ensuring that the brief stays on the user's machine. An open ONNX model allows for flexibility if the developer later decides to swap checkpoints without rewriting the page. This approach addresses the limitation of closed APIs, which send client briefs to someone else's server, potentially changing the cost of each call.
The model's accuracy on certain requirements has proven useful. For instance, it correctly identified a requirement for a video in a brief that initially received a 0% score for that criterion. However, the model can miss some requirements, such as a forge account, which is why plain-text flags are added to catch those cases.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.