Model Online / On-Site
Duration
Modules 5 modules

Course Objective & Scope

Standard PDF parsers mangle multi-column layouts, merged-cell financial tables, handwritten annotations and diagrams—causing immediate hallucinations in downstream RAG and document automation systems. This technical training focuses specifically on the ingestion and parsing layer that turns raw document, audio and image bytes into clean, structured Markdown and JSON.

Layout detection with Docling, Marker and modern OCR engines, resolving tables and charts with vision-language models (Qwen-VL, Gemini, GPT-4o), multi-speaker audio transcription with Whisper and pyannote, and extraction benchmarks (WER, TEDS) are covered hands-on. Participants build a high-accuracy ingestion module tested against their organisation's hardest documents and media files.

Who Should Attend?

  • AI engineers building intelligent document processing (IDP), RAG and multimodal applications
  • Data and software engineers digitising archives, contracts, invoices and call-centre recordings
  • Data scientists extracting structured information from documents and media
  • Prerequisite: Python programming and familiarity with Python libraries and APIs

Key Takeaways

  • Select the right extraction path (native parser, OCR or VLM) per document or page complexity
  • Convert complex tables, footnotes and multi-column layouts into Markdown/HTML/JSON without losing reading order
  • Apply audio preprocessing, VAD, speaker diarization and domain-jargon correction on noisy recordings
  • Turn embedded charts, diagrams and photos into structured descriptions using vision-language models
  • Measure extraction quality with automated metrics and route low-confidence pages to human-in-the-loop review

Curriculum

Document Layout Analysis and Modern OCR
  • Pitfalls of naive PDF text extraction, bounding-box detection and reading order
  • Extracting headings, paragraphs, lists and forms with Docling, Marker and Surya/PaddleOCR
Complex Table and Visual Content Extraction
  • Reconstructing merged-cell and multi-page tables in HTML/Markdown
  • Turning charts, flowcharts and technical drawings into structured JSON with VLMs
Audio Ingestion: ASR, VAD and Speaker Diarization
  • Audio normalization, Voice Activity Detection (VAD) and Whisper/faster-whisper
  • Speaker diarization with pyannote and domain-vocabulary correction
Hybrid Page Routing and Cost/Latency Optimization
  • Routing simple pages to fast parsers and complex pages to VLMs
  • Batch execution, caching and page-level parallelism
Extraction Quality Benchmarking and Human-in-the-Loop
  • Table (TEDS), text (CER/WER) and schema validation metrics
  • Confidence scoring and review queues for ambiguous extractions

Other Courses in This Category

  • dbt and Analytics Engineering Transformation Workflows

    Modular SQL transformations with dbt Core/Cloud, staging-intermediate-mart layers, Jinja macros, incremental models, snapshots, testing and CI/CD workflows for data and analytics engineers.

    Data Science & AI-Ready Data Engineering Corporate
    Live Course
    5 modules
    View
  • Data and AI Pipeline Orchestration with Apache Airflow

    Authoring production DAGs with Apache Airflow: TaskFlow API, dynamic task mapping, dataset/asset-aware scheduling, isolated execution on Kubernetes, error handling and AI/RAG pipeline orchestration.

    Data Science & AI-Ready Data Engineering Corporate
    Live Course
    5 modules
    View
  • Dirty Data, Clean CV: Real-World ML, Interviews & Career Dirty Data, Clean CV

    Advanced workshop on applying regression to messy real-world data, presenting model metrics to non-technical stakeholders, acing ML technical interviews, and building an impactful Data Science CV and GitHub portfolio.

    Data Science & AI-Ready Data Engineering Individual Workshop
    Live Course
    4 modules
    View
  • Data Lakehouse Architecture with Apache Iceberg

    Warehouse reliability on object storage: Apache Iceberg table internals, metadata tree, hidden partitioning, time travel, REST Catalog, multi-engine querying (Trino, Spark, DuckDB) and table maintenance.

    Data Science & AI-Ready Data Engineering Corporate
    Live Course
    5 modules
    View
  • Semantic Layer Design for BI, Text-to-SQL and AI Agents

    Centralising enterprise metrics and dimensions with dbt Semantic Layer (MetricFlow) and Cube so BI tools, Text-to-SQL systems and analytical AI agents return consistent numbers.

    Data Science & AI-Ready Data Engineering Corporate
    Live Course
    5 modules
    View
  • Metadata Engineering for RAG and Autonomous Agents

    Metadata engineering to boost retrieval precision and security in RAG and agent systems: schema design, automated extraction, pre-filtering, temporal validity and ACL permission filtering.

    Data Science & AI-Ready Data Engineering Corporate
    Live Course
    5 modules
    View

Get in touch