PDF to JSON with Local LLM

I am looking for an experienced Python developer to build a private, local document processing pipeline. The goal is to extract specific data points from variable-layout text PDFs and save them cleanly into a database via a structured schema. Requirements & Technical Constraints: No Third-Party APIs: We cannot use OpenAI, Gemini, Claude, or any external APIs due to strict data privacy rules. The entire system must run completely offline on a local server. Tech Stack: Python, PyMuPDF (or pdfplumber), and Ollama running a local model (such as Qwen2.5-7B or Llama-3.1-8B). Variable Layouts: The documents have similar text data but the visual layouts, formatting, and specific wording change significantly from file to file. Traditional coordinate-based PDF parsing will not work. Structured Output: The local LLM must be configured using Pydantic or Ollama’s structured JSON output feature to guarantee a strict schema response. Validation Layer: A Python post-processing layer needs to run a fast local fuzzy-matching script (RapidFuzz or TheFuzz) to map extracted text entities back to a Master database reference table. Scope of Work: Text Extraction Script: Read digitally selectable PDFs efficiently. Local LLM Integration: Prompt engineering and schema enforcement via Ollama to return a clean JSON payload. Fuzzy Database Reconciliation: Compare extracted values with a master reference list to correct minor spelling or naming variants automatically. FastAPI Wrapper: Wrap the working pipeline into a simple API endpoint (POST /parse-document) so our frontend can submit files and receive structured data back in 2-3 seconds. Deliverables: Clean, commented Python codebase. A [login to view URL] file detailing necessary packages. A brief setup guide explaining how to deploy the script, database connection string configuration, and Ollama configuration on an Ubuntu instance equipped with an NVIDIA GPU (e.g., A4000). More details along with pdf will me shared once finalised the service provider.

Project ID: 40523193