Advanced Knowledge & Agentic Systems with LLMs
TRAINING DESCRIPTION
This training is the next step after Buzz-Free Programming with LLMs. It moves beyond basic retrieval into the techniques that make LLM-based knowledge systems work in production: advanced and modular RAG, metadata-driven knowledge bases, graph-based retrieval, and self-maintaining knowledge layers. The second half focuses heavily on agents — frameworks, patterns, workflows, and deep agents that solve complex domain problems. As in the previous edition, the approach stays pragmatic and buzz-free: knowing when a technique earns its complexity, and when it doesn’t.
Order dedicated trainingBASIC PROGRAM
- Module 1: Advanced and Modular RAG
- Module 2: Metadata and Knowledge Base Design
- Module 3: Agentic RAG
- Module 4: Graph RAG
- Module 5: Compounding Knowledge: LLM-Wiki and Curation
- Module 6: Agentic Frameworks and Patterns
- Module 7: Agentic Workflows and Deep Agents
DETAILED PROGRAM
Module 1: Advanced and Modular RAG
- Recognizing where naive RAG hits its ceiling in production
- Pre-retrieval optimization: query rewriting, routing, intent classification, HyDE, step-back prompting
- Post-retrieval optimization: cross-encoder reranking, context compression, deduplication
- Context engineering at indexing time rather than query time
- Modular, composable pipelines: retrievers, rerankers, routers, and validators as swappable components
- Self-correcting retrieval loops (Self-RAG)
- Multimodal and visual document retrieval for scientific and technical documents
Module 2: Metadata and Knowledge Base Design
- Treating metadata as a context layer, not an afterthought
- LLM-generated metadata: automated extraction of content, technical, and semantic annotations at ingestion
- Metadata filtering: non-semantic pre-filtering by type, version, and access before vector search
- Multi-hop and multi-meta filtering for complex queries
- Provenance and lineage: source authority, versioning, and chain of custody
- Governance: access control, freshness and quality scoring, multi-source conflicts
- Structuring knowledge bases from messy organizational data
Module 3: Agentic RAG
- Adding a reasoning loop around retrieval: deciding when to retrieve, reformulate, or stop
- Query decomposition and parallel sub-queries
- Self-validation and iterative refinement of retrieved context
- Single-agent versus multi-agent retrieval architectures
- Knowing when agentic retrieval is worth the added latency and cost
Module 4: Graph RAG
- Moving beyond vectors: relationships, multi-hop, and entity-centric questions
- Building and querying knowledge graphs alongside vector stores
- Hybrid graph and vector retrieval
- Trade-offs of Graph RAG: cost, maintenance, and when it actually pays off
Module 5: Compounding Knowledge: LLM-Wiki and Curation
- Shifting from query-time retrieval to ingestion-time knowledge building (the LLM-Wiki pattern)
- Architecture: immutable sources, derived wiki, and the schema file
- The lint step: detecting contradictions, stale information, and orphan pages
- Time-awareness: distinguishing true-but-outdated knowledge from current knowledge
- Knowledge curation as a continuous, model-maintained process
- Enterprise reality check: scale, concurrency, and the limits of compiled knowledge layers
Module 6: Agentic Frameworks and Patterns
- Core agentic patterns: tool use, planning, reflection, and multi-agent collaboration
- The framework landscape and how to choose between options
- Type-safe, async-first agent design with Pydantic AI
- Knowing when a framework adds real value and when it is overhead
Module 7: Agentic Workflows and Deep Agents
- Workflows versus autonomous agents: predefined orchestration versus model-driven control
- Deep agents: planning loops, file operations, sub-agent delegation, and sandboxed code execution
- Human-in-the-loop approval in long-running tasks
- Building with Deep Agents and Pydantic-Deep
- Solving complex domain problems through decomposition and self-correction
KEY TAKEAWAYS
- Move beyond naive RAG to advanced, modular, and graph-based retrieval architectures.
- Design metadata and knowledge bases that make retrieval reliable, governed, and time-aware.
- Build self-maintaining knowledge layers that compound over time instead of rediscovering knowledge on every query.
- Master agentic patterns, frameworks, workflows, and deep agents for complex domain problems.