Hackathons: Kaggle (page 2)
Hackathons · Kaggle: 157 entries in total; this page lists entries 16–30.
Gem-Care (Gemma 4 Good Hackathon)
Evidence grade: for discovery only; must not by itself support a formal conclusion.
Gemma-4-Enriched with Multimodal Clinical-context Adaptation for Recognition Enhancement of Non-Normative Speech
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
TrueVoice: AI Voice Deepfake Detector
Evidence grade: for discovery only; must not by itself support a formal conclusion.
Protecting people from deepfake phone scams using Gemma 4's multimodal audio encoder
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
KawanIsyarat: Offline Gemma 4 for BISINDO Communication
Evidence grade: for discovery only; must not by itself support a formal conclusion.
A local-first Android app helping Deaf BISINDO users and hearing Indonesians communicate offline
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
Gilbeot(길벗)
Evidence grade: for discovery only; must not by itself support a formal conclusion.
On-device walking assistant for elderly users
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
PreVillage
Evidence grade: for discovery only; must not by itself support a formal conclusion.
A source-backed navigator for Nepal’s government services, built to find the office route, not just the form.
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
PathOS : Open Source AI for Histopathology Screening
Evidence grade: for discovery only; must not by itself support a formal conclusion.
Democratizing the diagnostic microscope. AI histopathology screening for hospitals, researchers, and places specialists never reach
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
MindBridge
Evidence grade: for discovery only; must not by itself support a formal conclusion.
A Hindi-First, Offline PHQ-9 / GAD-7 Screening App for India's ASHA Workforce, Built on Gemma 4 E2B + Cactus + Unsloth fine-tune
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
GrammarGym: fake language learning as a benchmark
Evidence grade: for discovery only; must not by itself support a formal conclusion.
Probe LLM in-context learning abilities using artificial grammars and fake creole emergence
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
EphLangBench: In-Context Learning Beyond Memorization
Evidence grade: for discovery only; must not by itself support a formal conclusion.
Evaluating LLM learning, cognitive flexibility and error correction by forcing models to solve algorithms in never-seen-before syntax.
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
ESFP Benchmark: Epistemic Stance Flexibility Probing in Large Language Models
Evidence grade: for discovery only; must not by itself support a formal conclusion.
A metacognition benchmark that tests whether LLMs can activate and suppress their own epistemic subjectivity w.r.t. how they are prompted
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
MetaBound: Can AI Models Map the Boundaries of Their Own Competence?
Evidence grade: for discovery only; must not by itself support a formal conclusion.
A Benchmark for Metacognitive Boundary Awareness in Large Language Models.
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
RIAC — Repetition-Induced Attention Collapse
Evidence grade: for discovery only; must not by itself support a formal conclusion.
100 distractors are easier to ignore than 1. We found the exact load level where frontier LLMs lose selective attention — and why.
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
ABC: Attention Benchmark for Cognition
Evidence grade: for discovery only; must not by itself support a formal conclusion.
Beyond local matching: evaluating selective attention through structure, grouping, and distraction
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
Turn Bench
Evidence grade: for discovery only; must not by itself support a formal conclusion.
A collection of turn-based games to test executive functions
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24
SecureExec-Bench: Measuring Executive Function Security in Large Language Models
Evidence grade: for discovery only; must not by itself support a formal conclusion.
Can frontier LLMs protect secrets under pressure? A 5-task benchmark of security-relevant executive functions across 14 models
Evidence use: For discovering how winning projects were implemented, their data resources, evaluation methods and application constraints; specific conclusions must be checked against the original project.
Limitations: Hackathon awards are not equivalent to peer review or independent reproduction. Project descriptions may change with the Kaggle page; check code, data and dependency versions before use.
Link last checked: 2026-09-24