Ali Ghazanfar@ali_ghazanfar544
🤝 Top Collaborators
🤓 Latest Submissions
UNIFY SUITE: AI workforce on Vultr
UNIFY SUITE is an AI workforce for the five departments most companies run on: Sales, Finance, HR, Support and Procurement. Each gets its own agent. The Lead Qualifier rates leads and cites its evidence, the Invoice Processor reads supplier invoices, the Resume Screener quotes the resume for every skill it credits, the Ticket Investigator checks a customer's real orders, and the Vendor Comparator explains a ranking that code computes. Aura, a chat assistant, routes plain-English requests to the right agent and can only do what the signed-in user may do. The rule is simple: AI reads, code decides, humans approve. The model only extracts, classifies and drafts, and every reply must pass a schema. Matching, ranking, duplicate checks and permissions are plain, tested Python. Payments, refunds and purchase orders wait in an Approval Center for a manager. The main flow is supplier payment verification. A clerk uploads an invoice, the AI reads it, code matches it to the purchase order and blocks duplicates, and the run pauses for a finance manager. Approve, and a payment is created and verified. Reject, and verification proves no money moved. The Command Center logs every agent and tool call, and the audit log is append-only. Every AI result has a Why? panel that shows what the AI read and what code decided. Drafted emails pass a quality gate in code, so a reminder can never quote an amount that is not in the facts. Hidden instructions in an invoice or call transcript are flagged to the manager. Everything runs on Vultr. One Cloud Compute VM hosts the app (React, FastAPI and PostgreSQL in Docker), Coolify from the Vultr Marketplace deploys it from GitHub with HTTPS, and every agent uses glm-5.3-flash through Vultr Serverless Inference. On a fictional data set with an answer key, 57 of 57 invoices were handled correctly, and 406 automated tests cover the approve and reject paths.
8 Nov 2026
