Scoring version lexicon-v1
Scoring Lab
Experimental baseline. Type any title and description to see how the same rules applied to every gallery project produce the Experimental Description-Based Indicator. It counts selected terminology; it is not an official judging score, a quality verdict or a winner prediction. The four criteria and their equal weighting come from the hackathon's official rules (section 7); the indicator itself is this tool's own description-based proxy.
Changing these weights changes how the available description evidence is prioritized. It does not change the projects or verify their actual performance.
Total 100%. Moving one slider keeps your value exactly and shares the remainder among the other three in proportion to their current values.
= (65×25 + 65×25 + 50×25) / 75
Rules
Input. Only the project title and its short gallery description. Award or winner labels, likes, comments and team members are never used.
Indicator per criterion (0–100). Each criterion has a fixed list of textual cues (right). Each cue counts once. Distinct cues found → indicator: 1 → 50, 2 → 65, 3 → 80, 4 or more → 95. This measures how much the description states about a criterion, not how good the project is.
Insufficient Evidence. If no cue is found for a criterion, the indicator is left empty. It is never set to 0 or to a neutral value.
Not scorable. If the description has fewer than 5 words, or fewer than 2 criteria have evidence, no weighted score is produced. The project stays searchable.
Weighted score. Σ(indicator × weight) ÷ Σ(weight), over criteria that have evidence. With all four present this is exactly (T×wT + D×wD + I×wI + Q×wQ) ÷ 100. Example: 80, 70, 90, 85 at 25% each gives 81.25. The share of weight that had evidence is shown next to every score.
Weights. Always total 100%. Moving one slider keeps your chosen value; the remainder is shared among the other three in proportion to their current values, rounded so the total stays exactly 100.
Reproducibility. No randomness and no AI calls. The same dataset, scoring version and weights always give the same ranking. Ties are broken by Devpost project ID.
Similar projects. Word overlap (TF-IDF cosine) between titles and descriptions.
Cue lists
- AI / model use ai|gpt|llm|language models?|machine learning|neural|openai|codex
- agents / automation agents?|agentic|automat…|workflows?
- speech / vision / multimodal voice|speech|vision|image recognition|multimodal|camera|ocr
- retrieval / data rag|retrieval|embeddings?|vector|knowledge graph|database|dataset
- real-time / on-device real[- ]?time|offline|on[- ]device|edge|low[- ]latency
- platform / integration api|apis|sdk|integrat…|plugin|extension|open[- ]source|pipeline
- interface / UX interfaces?|ui|ux|user experience|dashboard
- ease of use intuitive|easy|effortless|simple|seamless|frictionless|one[- ]click
- accessible design accessible|accessibility|inclusive design|screen reader
- interactive / conversational interactive|conversational|chat…|gamif…|visuali[sz]…
- personalisation personali[sz]…|tailored|adaptive|customi[sz]…
- health / wellbeing health…|medical|patients?|mental|wellbeing|well-being|clinic…|therapy
- education educat…|students?|learn…|teach…|school|tutor…|literacy
- inclusion / access disabilit…|disabled|blind|deaf|elderly|seniors?|underserved|low[- ]income|rural|language barriers?|multilingual|inclusi…
- environment climate|sustainab…|carbon|energy|environment…|waste|agricultur…|farmers?
- safety / civic safety|emergenc…|disaster|crisis|fraud|scam|legal|civic|nonprofit|communit…|government
- scale / economic millions?|billions?|global|small business…|jobs?|economic|poverty|affordable
- novelty claim first|novel|never before|new way|unique|innovative|breakthrough
- reframing reimagin…|rethink…|transforms?|turns?|instead of|unlike
- bridging / combination bridg…|combin…|brings? together|connects?|fusion
- specific problem framing so that|helps? w+ w+|without (?:needing|having)|in seconds|in minutes