DevPiggy 🐷
8,011 projects. One dataset. Three search methods. Different answers.
An independent PIGI research demonstration by FreeIntelligence.Institute.
Explore how search methods and analytical assumptions influence what you find in a large public dataset. This is not official judging and does not identify winners.
Deterministic full-text search, case-insensitive, with word stemming. Quotes for phrases, minus to exclude. No AI. Leave empty to rank everything.
Changing these weights changes how the available description evidence is prioritized. It does not change the projects or verify their actual performance. Used only when you sort by the experimental indicator.
Total 100%. Moving one slider keeps your value exactly and shares the remainder among the other three in proportion to their current values.
Search relevance answers “does this relate to my query?”. The Experimental Description-Based Indicator (Keyword Baseline v1) counts selected terminology in the gallery description; it is not an official judging score and 95 does not mean a project is technically excellent. They are separate numbers. The four criteria and their equal weighting come from the official rules; the indicator does not. Competition categories are unclassified: the gallery does not publish them.
Loading indexed projects…
Why we built DevPiggy
DevPiggy was inspired by the results of a recent OpenAI hackathon and public discussions questioning how evaluation criteria and project selection are understood. These discussions raised a broader research question:
How much can a result change when the underlying dataset remains the same, but the search method, visibility, or analytical assumptions change?
We explored this question using publicly available AI systems and observed that different systems reported different source-access limitations. Rather than attempting to declare alternative winners, we built a practical demonstration using the publicly accessible Devpost gallery. DevPiggy lets you explore how Keyword, Semantic and Hybrid retrieval expose different projects from the same dataset.
What DevPiggy demonstrates
- Dataset visibility affects conclusions.
- Exact keyword matching can miss conceptually relevant information.
- Semantic retrieval can find related meanings but may introduce false positives.
- Hybrid retrieval combines different signals.
- A result can be computationally consistent while still being incomplete or misleading.
- Transparent methods and source links help users understand how results were produced.
What DevPiggy does not claim
- It does not determine who should have won the hackathon.
- It does not accuse organizers or participants of wrongdoing.
- It does not verify the technical functionality of every submitted project.
- It does not reproduce the official judging process.
- It does not claim access to private judging information.
- It does not prove that a particular contest outcome was incorrect.
The goal is to demonstrate information-processing mechanisms, not to issue a verdict about the competition.
Beyond hackathons
The same information-processing challenges occur in:
- ESG reporting and evidence verification
- BIM/CAD technical documentation
- Business intelligence
- Recruitment and application screening
- Research and knowledge discovery
- Cybersecurity and threat detection
- AI output evaluation and auditability
DevPiggy is a small public demonstration of principles used in PIGI's broader research and development. It does not on its own validate every PIGI capability.