Tutorials¶
Interactive Jupyter notebooks covering Director-AI from first principles to production deployment. Every notebook runs in Google Colab with zero local setup.
Learning Path¶
graph LR
A[00 Quickstart] --> B[01 Coherence Engine]
B --> C[09 Production Guardrails]
B --> D[02 Streaming Oversight]
C --> E[10 Vector RAG]
C --> F[12 Domain Presets]
D --> G[11 Streaming Deep Dive]
E --> H[06 Medical RAG]
F --> I[13 Batch Processing]
I --> J[14 Enterprise]
J --> K[15 Fine-Tuning]
Getting Started¶
Start here. These two notebooks teach the core concepts in under 30 minutes.
| # | Notebook | What You Learn | Time | Colab |
|---|---|---|---|---|
| 00 | Quickstart | Install, score, guard, stream, presets | 5 min | |
| 01 | Coherence Engine | CoherenceScorer, SafetyKernel, CoherenceAgent, dual-entropy formula | 15 min |
Core Features¶
Deep dives into the four pillars of Director-AI.
| # | Notebook | What You Learn | Time | Colab |
|---|---|---|---|---|
| 09 | Production Guardrails | guard() for OpenAI / Anthropic / Bedrock / Gemini / Cohere, failure modes, streaming guards |
20 min | |
| 10 | Vector RAG Pipeline | Semantic retrieval, ChromaDB, pluggable backends, reranking, multi-tenant KB | 25 min | |
| 11 | Streaming Halt Deep Dive | Hard limit, sliding window, trend detection, async, per-token visualization | 20 min | |
| 12 | Domain Presets & Config | 8 profiles, env vars, YAML, backends, strict mode, multi-GPU, LLM-as-judge | 15 min |
Domain Applications¶
Real-world integrations and domain-specific patterns.
| # | Notebook | What You Learn | Time | Colab |
|---|---|---|---|---|
| 02 | Streaming Oversight | StreamingKernel basics, token-by-token monitoring | 10 min | |
| 03 | Vector Store | VectorGroundTruthStore, InMemoryBackend, fact ingestion | 10 min | |
| 05 | SSGF Geometry | Self-similar geometry foundation concepts | 10 min | |
| 06 | Medical RAG Chatbot | Healthcare-specific guardrails, high thresholds, evidence citations | 20 min | |
| 07 | LangChain Integration | CoherenceCallbackHandler, chain integration, output parsing | 15 min | |
| 08 | Provider Adapters | OpenAI, Anthropic, Bedrock, Gemini, Cohere adapter patterns | 10 min |
Enterprise & Production¶
Scale, evaluate, fine-tune, and deploy.
| # | Notebook | What You Learn | Time | Colab |
|---|---|---|---|---|
| 04 | End-to-End Benchmark | Full benchmark suite, latency profiling, accuracy metrics | 15 min | |
| 13 | Batch Processing & Evaluation | BatchProcessor, evaluation pipelines, claim attribution, regression gates | 20 min | |
| 14 | Enterprise Multi-Tenant | Tenant isolation, REST/gRPC servers, Docker, Kubernetes, monitoring | 25 min | |
| 15 | Custom Fine-Tuning | JSONL data prep, validation, training, anti-forgetting, ONNX export, REST API | 30 min |
Prerequisites¶
All notebooks run on Python 3.11+ with pip install director-ai.
Notebooks requiring optional extras note this in their first cell:
| Extra | Install | Notebooks |
|---|---|---|
| NLI scoring | pip install director-ai[nli] |
01, 04, 06, 09–15 |
| Vector store | pip install director-ai[vector] |
10, 14 |
| Fine-tuning | pip install director-ai[finetune] |
15 |
| Server | pip install director-ai[server] |
14 |
| gRPC | pip install director-ai[grpc] |
14 |