Exam Parser Pipeline
Production-Grade PDF ParsingProduction-grade ingestion pipeline that parses competitive exam PDFs into structured, validated JSON data. Employs a robust multi-stage flow with automated visual/text segmenting, self-correcting Pydantic validation, and bounding-box diagram extraction.
Architecture Snapshot
- 9-stage sequential pipeline architecture isolates AI logic to a single parser stage.
- Self-correction loop catches Pydantic validation errors and requests LLM retries with traceback.
- Union merge coordinate cropping extracts visual math, chemistry, and diagram sub-regions.
- Factory provider layer decouples system code from LLM vendors (Gemini, Qwen, DeepSeek, local).