Ponti
An AI agent that verifies whether a creator's content matches a brand's campaign brief — approve or revise, in seconds, with reasons.
视频
技术栈
描述
Ponti Verify checks whether a creator's submitted content satisfies a brand's campaign brief. Send it the brief (key message, things to avoid, product URL) plus the creator's submission (caption + live post link), and it returns an APPROVED / REVISION_NEEDED / FLAG decision with a 0–100 confidence score and specific, human-readable reasoning — never a black box.
Use case: Africa's creator economy runs on brands manually reviewing dozens of creator submissions against briefs, and creators getting ghosted over vague feedback. Ponti Verify turns that check into an instant, auditable decision — so brands only pay for on-brief work and creators get fast, fair, specific feedback. A guardrail ensures anything under 70% confidence never auto-approves; it routes to a human. The AI is never the final authority over payment.
Who it's for
Brands, agencies, and creator marketplaces that pay creators for briefed content.
How it works / demo walkthrough
A caller sends the brief + submission to the agent's endpoint; the agent (Llama 3.3 70B) evaluates the match and returns a structured decision, confidence score, and actionable notes. Pay-per-call via OKX's x402 on X Layer. Demo in the X post shows a weak submission caught and revised, then an improved one approved.
OKX.AI listing / Agent details
Agent: Ponti Verify · Agent ID #6037 · X Layer (chainIndex 196)
Service: Content-Brief Match Check (A2MCP) · pay-per-call, 0.1 USDT
Endpoint: https://ponti-verify.onrender.com/verify
Owner wallet: 0x0d4fc50045c2f87a2783a5b9d84ce93304fbc2f3
本次黑客松进展
During the hackathon we built Ponti Verify from scratch: an AI agent that checks whether a creator's content matches a brand's campaign brief and returns an auditable approve/revise decision with a confidence score and specific reasoning (Llama 3.3 70B). We deployed it as a live, paid Agent Service Provider on OKX.AI — pay-per-call via x402 on X Layer (agent #6037), with the payment layer verified end-to-end (HTTP 402 → pay → 200). We also built a live product demo of the verification flow and a guardrail where any decision under 70% confidence routes to a human, never auto-approving payment.
融资状态
Not currently raising — bootstrapped, pre-revenue MVP built during the hackathon.