
The SME Loan Risk Meta-Validator is an end-to-end intelligent workflow designed to streamline and modernize SME credit assessment. It automatically ingests loan applications from spreadsheets, portals, and supporting documents, then extracts and normalizes key information using OCR and AI-driven field extraction. The system enriches each application with external credit bureau and sanctions screening data, evaluates data quality and consistency, and applies transparent rule-based checks to enforce lending policies. For eligible cases, an AI risk model produces nuanced risk and eligibility scores, which are combined with rules to determine whether to auto-approve, auto-reject, or escalate for human review. An agentic policy reviewer ensures alignment with credit policies, fairness guidelines, and regulatory expectations. Every application generates a complete audit artifact—including inputs, risk reasoning, rules fired, and review actions—providing full traceability for risk teams and regulators. This workflow improves decision accuracy, reduces manual workload, ensures compliance, and delivers a scalable, consistent framework for SME lending.
19 Nov 2025

"Speak & Send" is a full-stack, serverless application that makes global micropayments as simple as speaking. The Problem: Traditional online payments, especially for tipping creators or in-game purchases, are slow and clunky. They require navigating forms, entering credit card details, and are impossible to use in voice-first environments like VR/AR, smart speakers, or while driving. Our Solution: We built a "voice wallet" with a clean, modern frontend that features a "Hold to Talk" button. A user simply states their command, like "send 5 USDC to @coolstreamer." This triggers a complete AI-driven pipeline, all running on a globally fast Cloudflare Worker: Speech-to-Text: The user's audio is transcribed into text by the ElevenLabs API. Intent Parsing: This text is fed to a Cloudflare AI (Mistral) model, which intelligently parses the human-like phrase into a structured JSON command for the payment. Payment Execution: The system calls the Circle API to execute the payment in USDC on the Arc Testnet, a new L1 blockchain purpose-built for stablecoin finance. Voice Response: The success or error message is sent back to ElevenLabs and played as audio for the user, completing the "voice-in, voice-out" loop. This project is perfect for content creator tipping, in-game/metaverse purchases, and improving financial accessibility. It demonstrates a fast, secure, and incredibly human-friendly financial transaction.
8 Nov 2025

HoloMentorX is an AI-powered contract auditing and mentoring platform built for the Qubic Track, combining the speed of Groq, intelligence of Meta’s LLaMA, and modular automation via Fetch.ai agents. It enables developers to submit smart contracts and instantly receive deep audits, vulnerability analysis, human-readable explanations, and automated tests. Our architecture supports: Groq-accelerated LLaMA inference for fast, detailed contract breakdowns. Fetch.ai-based agents to modularize tasks like auditing, testing, and summarization. A clean, React-based frontend + FastAPI backend for seamless UX. Multi-agent reasoning: multiple LLM calls work together to offer an audit decision and next action. HoloMentorX aims to solve the bottleneck of slow, expensive audits—enabling safer Web3 development for all.
8 Jul 2025