Appearance
How Matching & AI Search Work
Veritybase connects corporate buyers and sustainability managers with verified solution providers using AI-assisted vector matching, semantic similarity, and dual search intent modes — while keeping greenwashing out of report audits.
1. From Need to Match
When you enter a search query (e.g. "We need to collect supplier carbon data and quantify Scope 3 value-chain emissions") or upload a sustainability report gap:
Semantic Vector Embedding: Veritybase converts your plain-text request into a 768-dimensional AI semantic vector representation (using Vertex AI Gemini embeddings). This captures deep conceptual intent beyond literal keyword matching.
Database Vector Cosine Search: The engine computes the Cosine Similarity ($\text{sim}$) between your query vector and 1,500+ provider solution vectors stored in PostgreSQL
pgvector.Composite Scoring & Trust Gating: Each candidate is scored using a composite Fit Score that balances semantic relevance with provider verification trust:
$$\text{Fit Score} = \frac{w_{\text{sim}} \cdot \text{sim} + w_{\text{trust}} \cdot \text{score}}{w_{\text{sim}} + w_{\text{trust}}}$$
- Match Score (0 - 100): Displayed on each result card to show overall fit.
- Trust Badge: Indicates whether claims are verified by third-party audit (
High,Medium,Low) or unverified.
2. Dual Search Intent Modes
Veritybase operates two distinct search intent modes tailored to your workflow:
| Search Mode | Primary Purpose | Trust Floor (min_trust_label) | Candidate Scope |
|---|---|---|---|
Ask Search (/solutions/ask) | Exploratory supplier discovery | unverified (rank 0) | Surfaces all relevant verified & unverified catalog solutions with clear badging so buyers find providers quickly. |
Report Audit Matching (/report/match_gaps) | CSRD / GRI audit compliance | low (rank 1) | Enforces strict audit-grade verification floors; unverified claims are 100% excluded to guarantee zero greenwashing. |
3. Automated PII & Privacy Safeguards
To comply with GDPR and global privacy standards, all search prompts automatically run through our PolicyGate redaction layer before reaching external AI models. Personal contact details (emails, phone numbers, SSNs, credit cards) are masked automatically while preserving search semantics.
4. Tips for Getting Better Matches
- Be Specific: "Reduce Scope 3 freight and logistics emissions" yields stronger vector alignment than "Be more sustainable".
- Look at Badges: A high Match Score with an Unverified badge indicates strong potential fit that requires supplier follow-up. A Verified badge indicates independently audited data.