LuzDeck: An AI-Powered Platform for Semantic Skill Matching and Verified Startup Collaboration

Abstract

Early-stage startup formation fails disproportionately due to inadequate team composition — a consequence of self-reported, unverifiable skills and keyword-based matching that cannot assess contribution depth or work-style compatibility. This paper presents LuzDeck, an AI- powered microservices platform that addresses these failures through four integrated components: (1) a Semantic Skill Matching Engine using Sentence-BERT embeddings and cosine similarity with a weighted composite score integrating semantic alignment, verified skill confidence, and contribution compatibility; (2) a real-time GitHub Verification Engine mining public repositories for a multi-factor Skill Confidence Score; (3) a Contribution Analytics Engine classifying developers into five behavioural archetypes via time-series commit statistics; and (4) an Intelligent Legal Automation module generating co-founder agreements in under 90 seconds from parameterised, lawyer-reviewed templates. Evaluated on 650 annotated developer profiles (Fleiss' κ = 0.83), the full system achieves 88.7% matching accuracy, outperforming keyword-only (63.0%), TF-IDF (71.2%), Word2Vec (76.4%), and Sentence-BERT-only (81.1%) baselines. A 50-founder pilot study demonstrates 41% reduction in candidate discovery time and 71% fewer manual verification steps compared to conventional platforms.