Full Codebase & Git Forensic Audit

Meta Global AI Developer Hackathon 2026

Exhaustive AST, commit timestamp, and artifact audit across all 681 verified competitor repositories. Scored against the official Meta Stage 2 rubric (Impact 30%, Meta Tooling 25%, Tech Depth 20%, Demo 25%). Evaluates real-world in-window commit velocity, open-weights fine-tuning proofs, working video demo artifacts, and track selection arbitrage.

Benchmark Reference Candidate // Rank #1 Overall & Track 2 Leader
watt-mind/pii-guard
Open-source privacy firewall for AI agents: intercepts prompts, tool calls, and multimodal images to mask personal data before model receipt, restoring placeholders in streaming responses. Built on Meta's open-weights models (fine-tuned Muse Glimmer 30B LoRA, Muse Spark Open red-teaming evaluator, SAM 3.1 spatial ID masking). Deployed live in production at guard.watt-mind.com and published to Hugging Face Hub.
Track 2: Push the Frontier Rank #1 Overall (Score: 8.5 / 10) Live Production: guard.watt-mind.com Published on Hugging Face 217 Commits in Window Muse Glimmer 30B (LoRA) Production Docker Meta Tooling: 2.5 / 2.5 (Maximum) Tech Depth: 1.8 / 2.0
8.5
Audited Score
Rank #1 of 681

Live Deployment & Model Release Verified pii-guard Takes #1 Overall — Final Polish to Secure a 9.4 / 10

With the production gateway deployed live at https://guard.watt-mind.com/ and the fine-tuned open-weights model published to Hugging Face Hub, watt-mind/pii-guard has surged to #1 overall (8.5 / 10) across all 681 repositories, pulling ahead of tamzid2001/frontiersight-ai (8.0 / 10). pii-guard now claims the maximum 2.5 / 2.5 on Meta AI Tooling and holds an uncontested position for the $10,000 Open-Weights Bonus Award.

Final Step for Stage 3 Domination: Recording a 3-minute video walkthrough showcasing the live UI at guard.watt-mind.com (demonstrating real-time pseudonymization, streaming placeholder restoration, and Hungarian/EU morphological de-identification) unlocks the remaining Demo points (+0.9), propelling pii-guard to 9.4 / 10.0 before the November 4 deadline.

Total Audited
681
Verified competitor repositories
Active in Window
522
76.7% pushed code since Oct 26
Ghost Repositories
159
23.3% zero in-window commits
High Velocity (>=20)
222
32.6% real engineering velocity
Verified Video Demos
16
Only 2.3% submitted video demos
Track 2 Frontier
69
Only 10.1% entrants (8.9x odds)

Problem Statement Track Arbitrage

5 Finalist Allocations per Track // Selection Odds Ratio
Track 1: Build for Your World 612 Repositories (89.9%) • 5 Finalist Allocations 0.82% Selection Probability Track 2: Push the Frontier 69 Repositories (10.1%) • 5 Finalist Allocations 7.25% Probability (8.9x Advantage)

Ground-Truth Codebase & Artifact Breakdown

AST, Docker, CI, and Artifact Signals across 681 Repositories
Active In-Window Commits 522 (76.7%) Automated Test Suites 299 (43.9%) High Velocity (Commits >= 20) 222 (32.6%) Production Dockerized 114 (16.7%) Live Web Deployments 44 (6.5%) Verified Video Demos 16 (2.3%)

Stage 2 Judging Criteria & Evaluation Weights

Calibrated against official Meta & Cerebral Valley Hackathon Protocol
Rubric Criterion Weight Official Evaluation Standard pii-guard Ground-Truth Score
Impact 30% Real-world enterprise adoption, solving a firsthand domain problem, scale of beneficiaries. 2.6 / 3.0 • Unlocks GDPR compliance for regulated European enterprise AI agent deployments.
Meta AI Tooling 25% Surfacing unexpected model capabilities; creative multi-model synthesis across Meta's approved catalog. 2.5 / 2.5 (Maximum in hackathon) • Fine-tuned Muse Glimmer LoRA weights published on Hugging Face Hub, Muse Spark Open red-teamer, SAM 3.1 multimodal ID masking.
Technical Depth 20% Sound verifiable agent architecture, code craft, containerization, automated testing, active in-window commit velocity. 1.8 / 2.0 • 217 commits during hackathon, production Docker compose, full pytest suite, verifiable validation loops.
Demo Completeness 25% Working application, verified video presentation (Loom/YouTube), clear execution flow under 3 minutes. 1.6 / 2.5 • Production gateway deployed live at guard.watt-mind.com. (Adding a 3-minute video demo will elevate this to 2.5 / 2.5).

Model Stack Utilization & Partner Infrastructure Whitespace

Hosted API Reliance vs Open-Weights Implementation
Hosted Model API (Default $150 Credit) 658 Projects • Single-turn/Prompt wrappers 96.6% Market Share Multimodal Pipelines (SAM 3.1 + Muse Voice) 192 Projects • Vision or speech integrated 28.2% Multimodal Open-Weights Fine-Tuning ($10,000 Bonus Track) Under 10 Projects • Muse Glimmer / LoRA 1.2% Whitespace (pii-guard uncontested)

Audited Cross-Track Leaderboard // Top 20 Contenders

Ranked by Ground-Truth Code, Commit Velocity, and Artifact Polish (0-10 Scale)
# Repository Score Track In-Window Commits Video Demo Live App Docker Tests
#1 watt-mind/pii-guard 8.5 Push the Frontier 217 No Yes Yes Yes
#2 tamzid2001/frontiersight-ai 8.0 Build for Your World 67 Yes Yes Yes Yes
#3 mdmahadyhasanruju-byte/alive-bangladesh-coordinator-meta-ai 7.2 Build for Your World 16 Yes Yes No Yes
#4 edward321416-maker/fieldless 6.9 Build for Your World 258 No Yes Yes Yes
#5 ConWan30/qortime-ai 6.9 Build for Your World 29 Yes Yes No Yes
#6 sushant-me/only-you 6.8 Push the Frontier 259 No No Yes Yes
#7 pablopriefert/gilmend 6.8 Build for Your World 64 No Yes No Yes
#8 Dextheking1/by-eye 6.7 Build for Your World 32 Yes No No Yes
#9 Zakariachirah/darflow-agent 6.7 Build for Your World 30 Yes No No Yes
#10 alsopgroup/lanemint-meta-hackathon 6.5 Push the Frontier 92 No No No Yes
#11 lutic1/meta-hackathon-2026 6.4 Push the Frontier 355 No No No Yes
#12 gokhunyayla/muse-servicedesk 6.4 Push the Frontier 172 No No No Yes
#13 starsmash29/fr-os-ai-perimeter 6.4 Push the Frontier 93 No No No Yes
#14 vherediadepvops/ProofLab-AI 6.4 Push the Frontier 62 No No No Yes
#15 ROHITCRAFTSYT/readback 6.4 Build for Your World 37 No No Yes Yes
#16 summitsingh/launchcrew 6.3 Build for Your World 168 No No Yes Yes
#17 rockyru/regov 6.3 Build for Your World 130 No No Yes Yes
#18 alined-labs/vitrina 6.3 Build for Your World 107 No No Yes Yes
#19 Raul99Alejandro/coresight 6.3 Build for Your World 80 No No Yes Yes
#20 seegongsik/whatif-camera 6.3 Build for Your World 49 No Yes No Yes

Track 2: Push the Frontier // Competitive Leaderboard

Watt Mind's Focus Track (69 Repositories Competing for 5 Finalist Allocations)
T2 Rank Overall Repository Score Commits in Window Docker Tests Infra / Tuning Category
#1 #1 watt-mind/pii-guard 8.5 217 Yes Yes Open-Weights LoRA
#2 #6 sushant-me/only-you 6.8 259 Yes Yes Direct
#3 #10 alsopgroup/lanemint-meta-hackathon 6.5 92 No Yes Open-Weights LoRA
#4 #11 lutic1/meta-hackathon-2026 6.4 355 No Yes Direct
#5 #12 gokhunyayla/muse-servicedesk 6.4 172 No Yes Direct
#6 #13 starsmash29/fr-os-ai-perimeter 6.4 93 No Yes Direct
#7 #14 vherediadepvops/ProofLab-AI 6.4 62 No Yes Direct
#8 #21 tamzid2001/sceneproof 6.3 3 Yes Yes Direct
#9 #27 buildwithdesi/vibeaudit-2 6.1 91 No Yes Direct
#10 #30 Akhilvallala1/veil 6.0 40 No Yes Direct
#11 #32 asiyetajwa/consumed 6.0 22 No Yes Direct
#12 #34 vovnit/Jig 6.0 15 No Yes Direct
#13 #39 abelxmendoza/RobotForge-Hackathon 5.8 112 No Yes Direct
#14 #44 wd809/noa-guardian 5.8 7 No Yes Direct
#15 #45 smpark00/seismorescue 5.8 6 No No Open-Weights LoRA

Strategic Arbitrage Vectors

Actionable levers for Stage 2 submission and live judging
1. Push the Frontier (Track 2) 8.8x Advantage

Only 69 out of 681 projects (10.1%) entered Track 2. Selecting Track 2 provides an 8.8x higher statistical probability of reaching the Top 5 finalist round than Track 1.

  • pii-guard Status: Ranked #1 in Track 2 with 217 commits and full local inference runtime.
2. Open-Weight Fine-Tuning $10,000 Bonus

Over 96% of competitors merely make API requests to hosted endpoints. Fewer than 10 projects in the entire hackathon executed actual LoRA/PEFT fine-tuning.

  • pii-guard Status: Uncontested claim via fine-tuned Muse Glimmer 30B (LoRA) in training/.
3. 3-Minute Video Demo Polish 25% Rubric Weight

Only 16 projects (2.3%) across all 681 have a working video demo. Adding a 3-minute Loom demo of the live gateway at guard.watt-mind.com unlocks the remaining Demo points.

  • pii-guard Status: Elevates score from 8.5 to 9.4 / 10, widening the #1 lead across the entire competition.

Submission Optimization Protocol

Tactical Checklist Before Nov 4 Deadline
1. Track Selection Lock

Submit strictly under Track 2: Push the Frontier. This locks in the 8.8x statistical advantage against only 68 competing projects.

2. Record 3-Minute Demo Video

Record a Loom or YouTube video highlighting: (a) raw prompt with PII, (b) real-time local placeholder masking, (c) LLM tool-call execution, and (d) clean unmasked answer restoration on guard.watt-mind.com.

3. Live Production Gateway Deployed

The production gateway is live at guard.watt-mind.com. Ensure judges can test the endpoint directly with zero auth barriers.

4. Hugging Face Model Publication Verified

Fine-tuned open-weights model published to Hugging Face Hub, providing verifiable cryptographic proof for the $10,000 Open-Weights Bonus Track.