Job Title: Senior AI / Python Engineer for Enterprise LLM Prototype (PDF Coordinate Mapping) Project Type: Fixed Price (Milestone 1) Budget: $1000 - $2000 (For this initial PoC milestone) ABOUT US We are an Abu Dhabi-based enterprise AI startup. We are building an "Agent-as-a-Service" infrastructure for corporate procurement and legal teams. We operate on a strict "Work for Hire" model. All code, architecture, and IP developed during this contract will be 100% owned by us. An NDA and IP Assignment Agreement is mandatory before beginning work. THE MISSION (MILESTONE 1) We need to build our core Proof of Concept (PoC): The Glass Box Trust Engine. Enterprise companies do not trust LLM hallucinations. We need a deterministic prototype that takes a business PDF, extracts data using an LLM, checks it against hardcoded rules, and visually proves the AI's logic by drawing bounding boxes on the original PDF. TECHNICAL REQUIREMENTS (The PoC Workflow) 1. Ingestion: A simple Python/Streamlit frontend where a user uploads a 3-page PDF (e.g., a Vendor Quote). 2. Extraction & Logic: Send the PDF text to the OpenAI API (GPT-4o). Have the LLM check the text against 3 specific hardcoded rules (e.g., "Pump pressure must be more than 120 Bar"). 3. Spatial Verification (The Hard Part): If the LLM finds a violation (e.g., the PDF says "110 Bar"), the system must use a library like PyMuPDF or pdfplumber to find the exact X,Y coordinates of that specific text string on the original PDF. 4. Output: The system outputs a new version of the PDF with a Red Bounding Box drawn around the violating sentence, alongside a simple Pass/Fail dashboard. SKILLS REQUIRED • Expert in Python. • Deep experience with OpenAI API, LangChain, or LlamaIndex. • Experience with complex PDF parsing and spatial mapping (PyMuPDF, fitz, OCR). • Ability to deploy a quick frontend using Streamlit or Gradio for us to test. HOW TO APPLY To prove you have read this brief and understand the technical challenge, please start your proposal with the word "DETERMINISTIC." In your proposal, please answer: 1. Have you ever mapped LLM text outputs back to specific X,Y coordinates on a source PDF? How would you approach this? 2. What is your estimated timeline to deliver this 1-feature PoC? If this PoC is successful, we have a 6-month roadmap of continuous, high-budget development to build out the full multi-agent architecture.