<p>Own a full procurement automation platform end to end — invoice intelligence, SAP Ariba RPA, ERP reconciliation, Teams bots, and AI document verification — for a multi-brand hotel group (New York, Spain, Dubai) and enterprise clients including Ariba, Microsoft, Google, and AON.</p><p> Built an LLM-vision invoice extraction pipeline (FastAPI + Google Gemini 2.5) reading invoices from email/SFTP/SharePoint across three hotel-brand markets, without hand-written OCR templates.</p><p> Reached a 93% straight-through validation rate on the Dubai deployment; the remaining ~7% are correctly routed to manual review (missing PO number, unclear scan, or GRN not yet posted) instead of being silently mis-processed.</p><p> Built a Playwright RPA bot that creates the invoice directly on the Ariba web UI once extraction and validation pass — the automated hand-off that closes the loop with no manual data entry.</p><p> Built a second bot creating invoices for both material-PO and service-PO line items (handled differently in Ariba), plus a spend-analysis module (pandas/statsmodels).</p><p> Built a Selenium bot suite covering the full service-entry-sheet (SES) lifecycle — create, reject, delete — and purchase-order change requests, all driven directly on the Ariba UI.</p><p> Built an Ariba-vs-SAP-S4/HANA validation service that checks POs before posting, preventing duplicate invoice submissions into SAP.</p><p> Designed a family of APScheduler jobs syncing PR, GRN, Invoice/IR/Payment, and SES data from Ariba into SQL Server on a recurring cadence, feeding reconciliation and reporting.</p><p> Built a unified Pulse Report dashboard consolidating PR, GRN, SES, and Invoice status into one view for procurement teams, with pandas/openpyxl Excel export.</p><p> Built two Bot Framework bots (Adaptive Cards, MSAL/Graph authentication) — one for PR creation/approval, one for IT ticketing — with role-aware views for business users, managers, and approvers.</p><p> Built a GPT-4o vision pipeline validating PAN, GST, MSME, and cancelled-cheque documents during supplier onboarding — classifying document type, extracting structured fields, flagging tampering, and fuzzy-matching against submitted registration data.</p><p> Automated Master Service Agreement generation for Ariba vendor onboarding (Flask, python-docx, docx2pdf), replacing manual contract drafting with a templated DOCX/PDF workflow.</p><p> Built and redeployed a LlamaIndex + ChromaDB retrieval assistant three times — an internal support bot, an Ariba documentation assistant, and a knowledge base for client Mead Johnson — with OCR fallback and gTTS voice output.</p><p> Containerized every service with Docker and deployed via GitHub Actions CI/CD to self-hosted runners across Azure and AWS (ARM64/Graviton for the Dubai deployment).</p>
