/ Home

Kactii GenAI Bootcamp - Season 4 / S4

Note: Military Style Gamified Learning Framework

Syllabus

Updated: March 1, 2025

Chapter 1: Introduction to GenAI Engineering Start your journey into GenAI engineering. Understand the key concepts and tools used in the industry. Gain a comprehensive overview of AI and machine learning technologies. Set the stage for advanced learning and practical application. Lay a strong foundation for your GenAI career.

Chapter 2: Large Language Models (LLM) Fundamentals Dive deep into Large Language Models including OpenAI GPT series, Claude (Anthropic), Gemini (Google), and Llama Family (Meta). Understand model architectures, training processes, and capabilities. Learn prompt engineering, fine-tuning techniques, and model selection strategies for different use cases.

Chapter 3: LangChain Ecosystem Mastery Master the complete LangChain ecosystem including LangChain core framework, LangGraph for graph-based workflows, LangFuse for observability and analytics, and LangSmith for development and debugging. Build complex AI applications with chain-of-thought reasoning and multi-step workflows.

Chapter 4: Vector Databases and Embeddings Master vector databases including Qdrant, Weaviate, FAISS, Milvus, and Chroma for AI-native applications. Learn embedding techniques, similarity search, and retrieval-augmented generation (RAG). Implement semantic search, document retrieval, and knowledge base systems for GenAI applications.

Chapter 5: Model Context Protocol (MCP) and Agent-to-Agent Communication Learn Model Context Protocol for standardized AI model communication and Agent-to-Agent (A2A) communication patterns. Implement multi-agent systems, orchestrate complex workflows, and build collaborative AI applications. Master inter-agent messaging, task delegation, and distributed AI processing.

Chapter 6: Local LLM Deployment with Ollama Master local LLM deployment using Ollama for privacy-focused and cost-effective AI applications. Learn model quantization, optimization techniques, and hardware requirements. Implement offline AI capabilities, custom model fine-tuning, and edge deployment strategies for production environments.

Chapter 7: HuggingFace Ecosystem and Model Hub Master the HuggingFace ecosystem for model deployment, sharing, and collaboration. Learn to use Transformers library, Datasets, and Spaces for building and deploying AI applications. Implement model fine-tuning, dataset preparation, and community collaboration through the HuggingFace platform.

Chapter 8: Database Technologies for GenAI Applications Master multiple database technologies including MongoDB, PostgreSQL with SQLAlchemy and Alembic, Supabase, and SQLiteDB. Learn to design data architectures for AI applications, implement database migrations, and optimize queries for large-scale GenAI systems with proper data modelling.

Chapter 9: Workflow Orchestration and Background Processing Master workflow orchestration using Prefect and background job processing with Dramatiq for scalable GenAI applications. Learn to design complex AI pipelines, handle asynchronous tasks, and implement distributed processing systems for production-grade AI workflows.

Chapter 10: Interactive Development with Notebook LLM and PyNotes Master interactive development environments using Notebook LLM for AI experimentation and PyNotes for technical documentation. Learn to create reproducible research notebooks, document AI experiments, and build interactive AI prototypes using modern notebook-based development workflows.

Chapter 11: Alternative GenAI Frameworks and Haystack Explore alternative GenAI frameworks beyond LangChain, focusing on Haystack NLP pipeline framework. Learn to build production-ready NLP applications, implement custom components, and compare different framework approaches for various GenAI use cases and requirements.

Chapter 12: Research Publication and Portfolio Development Complete your GenAI journey by publishing 1-3 research papers on ArXiv and building a comprehensive portfolio. Learn academic writing, research methodologies, and publication processes. Create projects for GitHub, GitBook documentation, Google Colab notebooks, and HuggingFace model sharing to showcase your expertise.


Weekly Sessions

Kactii GenAI Bootcamp S4 — 12-Week Master Plan

Model Provider: llama.cpp (local, no cloud API keys)
Hardware: RTX 3090 (24GB VRAM) + Ubuntu
Structure: 12 Chapters × 5 Sessions = 60 Sessions


Week 1 — GenAI Engineering Foundations

Chapter 1: Introduction to GenAI Engineering

Day Topic
1 Install llama.cpp from source, download first GGUF, run CLI inference
2 llama.cpp server mode, hit /v1/chat/completions via curl + Python
3 Tokenization hands-on with llama-tokenize, context window experiments
4 GGUF format deep dive, quantization types (Q4 vs Q5 vs Q8) comparison
5 Build reusable LlamaCppClient Python class used throughout the bootcamp

Week 2 — LLM Fundamentals

Chapter 2: Large Language Models (LLM) Fundamentals

Day Topic
6 Model families survey (GPT, Claude, Gemini, Llama) — run two GGUFs side-by-side
7 Zero-shot vs few-shot prompting harness in Python
8 Chain-of-thought + Jinja2 system prompt templating
9 LoRA/QLoRA concepts — compare base vs fine-tuned GGUF on domain task
10 Model selection mini project: pick the right GGUF for 3 different task briefs

Week 3 — LangChain Ecosystem

Chapter 3: LangChain Ecosystem Mastery

Day Topic
11 Wire LangChain ChatOpenAI to llama.cpp via base_url override
12 Build LCEL chains: prompt template → model → output parser
13 LangGraph intro — build a simple StateGraph with two nodes
14 LangSmith local tracing — log chain execution without cloud dependency
15 Multi-step workflow: LangGraph agent with conditional edges + llama.cpp brain

Week 4 — Vector Databases & Embeddings

Chapter 4: Vector Databases and Embeddings

Day Topic
16 Embedding concepts — use llama.cpp /v1/embeddings with a local embedding GGUF
17 FAISS hands-on: index documents, run similarity search manually
18 Qdrant via Docker — upsert vectors, metadata filtering, compare recall vs FAISS
19 Full RAG pipeline: loader → chunker → Qdrant retriever → llama.cpp LLM
20 RAG evaluation: chunk size tuning, hallucination on low-recall queries, mini project

Week 5 — MCP & Agent-to-Agent Communication

Chapter 5: Model Context Protocol (MCP) and Agent-to-Agent Communication

Day Topic
21 MCP protocol concepts — build a minimal MCP server with fastmcp
22 Register tools on MCP server, connect client to llama.cpp endpoint
23 Tool-call JSON parsing from raw GGUF output — structured outputs with Pydantic
24 A2A pattern: Orchestrator + Specialist agents communicating over HTTP
25 Multi-agent mini project: two llama.cpp-powered agents solve a task collaboratively

Week 6 — Local LLM Deployment & Optimization

Chapter 6: Local LLM Deployment with Ollama

Day Topic
26 llama.cpp GPU offloading — tune -ngl layers on RTX 3090, measure VRAM usage
27 Convert HuggingFace model to GGUF using convert_hf_to_gguf.py
28 Quantize to Q4/Q5/Q8 — benchmark tokens/sec and perplexity tradeoffs
29 Serve multiple models with llama-server — model switching strategies
30 Edge deployment simulation: run llama.cpp on CPU-only Ubuntu laptop, compare perf

Week 7 — HuggingFace Ecosystem

Chapter 7: HuggingFace Ecosystem and Model Hub

Day Topic
31 HuggingFace Hub tour — find, download, and convert models to GGUF locally
32 Transformers library: run inference locally, compare output vs llama.cpp GGUF
33 Dataset preparation with HuggingFace datasets for fine-tuning input
34 Fine-tune a small model with LoRA using peft + push adapter to HF Hub
35 Convert fine-tuned adapter to GGUF — serve it via llama.cpp, validate output

Week 8 — Database Technologies

Chapter 8: Database Technologies for GenAI Applications

Day Topic
36 MongoDB for chat history — store and retrieve conversation turns for llama.cpp context
37 PostgreSQL + SQLAlchemy: design schema for agent memory and task logs
38 Alembic migrations — version your DB schema as your agent app evolves
39 Supabase as a backend: vector extension + REST API for a RAG app
40 Multi-DB mini project: agent uses Mongo for memory, Postgres for structured logs

Week 9 — Workflow Orchestration

Chapter 9: Workflow Orchestration and Background Processing

Day Topic
41 Prefect intro — wrap RAG pipeline as a Prefect flow with tasks
42 Schedule and deploy a Prefect flow: nightly document re-indexing job
43 Dramatiq for async background jobs — trigger embedding on new file drop
44 Error handling and retries in Prefect + Dramatiq for flaky llama.cpp calls
45 End-to-end orchestrated pipeline: ingest → embed → query → log, fully automated

Week 10 — Interactive Development & Documentation

Chapter 10: Interactive Development with Notebook LLM and PyNotes

Day Topic
46 Jupyter + llama.cpp: build a reproducible experiment notebook
47 PyNotes for technical documentation — document your RAG and agent experiments
48 Notebook-driven prototyping: iterate on prompts and chains inside Jupyter
49 Build an interactive demo notebook: upload a PDF, query it via llama.cpp RAG
50 Portfolio notebook: clean, annotated, shareable on GitHub + Google Colab

Week 11 — Alternative Frameworks & Haystack

Chapter 11: Alternative GenAI Frameworks and Haystack

Day Topic
51 Haystack intro — OpenAIChatGenerator pointed at llama.cpp server
52 Build a Haystack document store + retriever + generator pipeline
53 Custom Haystack components — write a LlamaCppEmbedder component
54 LangChain vs Haystack side-by-side: same RAG task, both frameworks, compare DX
55 Haystack mini project: production-ready NLP pipeline with llama.cpp backend

Week 12 — Research, Portfolio & Capstone

Chapter 12: Research Publication and Portfolio Development

Day Topic
56 arXiv paper structure — abstract, methodology, experiments, results
57 Pick a paper topic from your bootcamp work (RAG eval, agent benchmarks, etc.)
58 Capstone build: LangGraph agent + Qdrant RAG + MCP tools + Mongo memory, all on llama.cpp
59 Deploy capstone as a FastAPI service — GitBook docs + HuggingFace model card
60 Demo day: present capstone + paper draft to cohort, peer review, submission prep

60 sessions · 12 weeks · 1 model provider — llama.cpp