March 19 — April 2, 2026. How building the product revealed the product's value proposition.
The MOTPE optimizer proved the knowledge graph has real predictive signal. 30 Optuna trials found a 12-parameter configuration that tripled accuracy. First out-of-sample validation: 84 predictions evaluated across 4 methods, 38 cumulative predictions scored over 2 weeks.
Full semantic layer operational. OWL ontology with 12 classes and 22 properties. SHACL validation. Nightly automated pipeline from web scraping through scoring, RDF conversion, triplestore loading, and prediction generation.
Stack-ranked candidate verticals. K-Pop scored 4.3/5 on data availability. Built complete OWL ontology with 8 classes and 10 edge types. First real weekly report: 8 entities scored. Same SBPI framework, different domain data.
The knowledge graph is not just infrastructure. It compounds weekly with every scoring cycle. Expert-designed ontology + SHACL validation + provenance tracking creates a defensible moat that commodity LLM outputs cannot replicate.
Multi-agent evaluation from three investor perspectives: YC judge, angel investor, institutional analyst. All critical fixes synthesized into v2. Use of Funds cost ontology with 6 cost types and hardware scaling model.
The last two weeks were not just deliverables shipped. Three parallel explorations — K-Pop vertical selection, autoresearch experiments, and YC AI agent stack ranking — each fed back into a single insight: the knowledge graph is an appreciating asset, and the process of building it IS the demonstration of its value.
Every milestone from March 19 through April 2. 20+ deliverables shipped across infrastructure, methodology, investor materials, cross-vertical expansion, and product development. Color-coded by project with deployment status.