Notebook Gallery¶

Use this gallery to choose a runnable Director-AI notebook by buyer problem, implementation track, and required optional extras. Each row links to the local notebook in the repository and the matching Google Colab launcher.

If you are evaluating commercial fit before opening a notebook, start with Applications and Market Map, then read Product Overview. If you are running a governed pilot, pair the notebook path with Evaluation Onboarding so the pilot produces useful evidence instead of a demo-only result.

Choose The Notebook By Outcome¶

Outcome	Notebook path
First guarded response	Quickstart, then Coherence Engine
Streaming halt proof	Streaming Oversight, then Streaming Halt Deep Dive
Private-fact grounding	Vector Store, then Vector RAG Pipeline
SDK or framework integration	Provider Adapters, LangChain Integration, Production Guardrails
Domain or regulated example	Medical RAG Chatbot, Domain Presets, Verification Gems
Enterprise pilot evidence	Batch Processing, Enterprise Multi-Tenant, Custom Fine-Tuning

Notebook-To-Market Map¶

Market question	Notebook path	Evidence produced
Can the guard reject a known false answer?	Quickstart, Coherence Engine	Approved/rejected pair with score and evidence
Can it protect streamed output?	Streaming Oversight, Streaming Halt Deep Dive	Halt metadata for a completed contradictory claim
Can it ground private knowledge?	Vector Store, Vector RAG Pipeline	Retrieved chunks tied to a verdict
Can it fit an existing stack?	Provider Adapters, LangChain Integration, Production Guardrails	Wrapped SDK/framework call path
Can it support a governed pilot?	Batch Processing, Enterprise Multi-Tenant, Custom Fine-Tuning	Threshold, tenant, deployment, and adaptation evidence

Track	Notebook	Audience	Use Case	Time	Extras	Colab
Foundations	Protect any LLM in 10 lines	First evaluation	Install Director-AI, understand the guarded workflow, score a response, wrap an SDK client, and inspect halt metadata.	5 min	base	Open
Foundations	Coherence Engine Quick Start	Technical evaluator	Understand CoherenceScorer, SafetyKernel, CoherenceAgent, and dual-entropy scoring.	15 min	nli	Open
Streaming Safety	Streaming Token-by-Token Oversight	Application engineer	Monitor streamed tokens and inspect contradiction-halt metadata before completion.	10 min	base	Open
Retrieval	Vector Store and Semantic Retrieval	RAG engineer	Load facts into a vector-backed ground-truth store and retrieve evidence for scoring.	10 min	vector	Open
Evaluation	End-to-End Guardrail Benchmark	ML evaluation lead	Run benchmark examples, profile latency, and interpret catch-rate metrics.	15 min	nli	Open
Domain Applications	Medical RAG Chatbot	Healthcare evaluator	Apply high-threshold medical guardrails with retrieval evidence and clinical disclaimers.	20 min	nli, vector	Open
Integrations	LangChain Integration	LangChain developer	Wire Director-AI into LangChain callbacks and chain output parsing.	15 min	langchain	Open
Integrations	Multi-Provider SDK Guard	Platform engineer	Compare commercial SDK adapters, cloud-runtime adapters, agent frameworks, Guardrails AI, and Vercel AI SDK adapter patterns.	10 min	base	Open
Production	Production Guardrails	Application engineer	Wrap production SDK calls, choose failure modes, and keep streamed-response checks evidence-bound.	20 min	nli	Open
Retrieval	Vector RAG Pipeline	RAG engineer	Build a semantic fact retrieval pipeline with ChromaDB, pluggable backends, reranking, and tenant-aware knowledge.	25 min	nli, vector	Open
Streaming Safety	Streaming Halt Deep Dive	Runtime engineer	Compare hard limits, sliding windows, trend detection, async streams, and per-token visualization.	20 min	base	Open
Configuration	Domain Presets and Configuration	Solutions engineer	Apply profiles, environment variables, YAML config, strict mode, multi-GPU settings, and LLM-as-judge options.	15 min	nli	Open
Evaluation	Batch Processing and Evaluation Pipelines	Evaluation engineer	Run batch scoring, claim attribution, dataset evaluation, and regression gates.	20 min	nli	Open
Enterprise	Enterprise Multi-Tenant Deployment	Enterprise platform team	Inspect tenant isolation, REST and gRPC services, Docker, Kubernetes, and monitoring patterns.	25 min	server, grpc, vector	Open
Model Adaptation	Custom Fine-Tuning	ML engineer	Prepare JSONL data, validate training examples, measure forgetting risk, export ONNX, and package held-out evidence before any tuning claim.	30 min	finetune, nli	Open
Verification	Verification Gems	Safety engineer	Exercise numeric, reasoning, temporal, consensus, conformal, feedback-loop, agentic, and REST verification modules.	15 min	base	Open
Demo	Streaming Halt Live Demo	First-touch evaluator	Open a focused Colab demo that shows opt-in streaming contradiction behavior without a local setup.	8 min	base	Open

Local Execution¶

pip install -e ".[dev,nli,vector,server]"
jupyter lab notebooks/

The manifest at notebooks/gallery.toml is the source of truth. The gallery validator fails if a notebook is added without a manifest entry or if this page omits a notebook link.