You can use this learning path with our AI sandbox. See instructions in #igiu-ai-learning slack channel.

Other resources of help:

1. Training Video
2. Powerpoint

The examples are based on Python 3.11+. Both Python scripts and Jupyter notebooks are available. We use UV for dependency management and execution.

1. Request access to sandbox #igiu-ai-learning.

   • Set up API keys and DB access based on the document here.
   • Update the sandbox.yaml file per your environment (OCI config, compartment, DB wallet path, bucket prefix, etc.).

2. Note the AI sandbox gives you access to two regions:

   1. Chicago: AI services, AI Playground, and Gen AI Agents are in this region.
   2. PHX: 23 AI Database, object buckets, and compute (if available) will be in this region.
   3. Object bucket in CHI is a read-only replica, so any files you drop in PHX will show up in Chicago automatically.
   4. Same bucket is shared by all users of AI sandbox; thus, best to use your Oracle User ID as prefix for your objects (configure it in sandbox.yaml).
   5. Same schema in 23ai will be shared by all users; thus, best to use your tables with your Oracle User ID as prefix for table names (configure in sandbox.yaml).
   • For DB examples: Download the 23ai wallet, unzip locally, and update sandbox.yaml with path, user, service name, password (via .env).

3. Setup UV environment:

   • Install UV: https://docs.astral.sh/uv/getting-started/installation/
   • uv sync to install dependencies (oci, oracledb, python-dotenv, etc.) from pyproject.toml.
   • Run uv run AISandboxEnvCheck.py to verify setup.
   • To run scripts: uv run <path/to/script> (e.g., uv run llm/cohere_chat.py).
   • For notebooks: Open in Jupyter/VS Code after uv sync.

4. We recommend following this progressive learning path, starting with core concepts and building to advanced integrations. Each module includes Python scripts, Jupyter notebooks, and detailed readmes with run instructions, docs, and project ideas.

   Core LLM and Agentic Basics:

   • llm: Introduction to OCI Generative AI APIs using Cohere and OpenAI-compatible models (chat, streaming, structured output, history, OCI Language NLP).
   • rag: Retrieval-Augmented Generation with OCI Agents, document citations, and home-grown pipelines using embeddings and Oracle DB.
   • function_calling: Function calling (tools) for single/multi-step agentic workflows with Cohere and Llama models, including classification.

   OCI AI Services:

   • speech: Text-to-Speech (TTS) and Speech-to-Text (STT) using OCI Speech with Oracle and Whisper models.
   • vision: Image/video analysis with OCI Vision (object/text detection) and Document Understanding (OCR, extraction), plus multimodal LLMs.
   • database: AI in Oracle Autonomous Database 23ai – NL2SQL, Select AI, database-centric RAG, semantic caching.

   Advanced: LangChain Integration (using LangChain 1.0.3):

   • langChain/llm: LLM interactions with chat, history, streaming, structured output, async, reasoning via OCI GenAI.
   • langChain/rag: RAG with document chunking, OCI embeddings, Oracle DB vector search, AIA reranking.
   • langChain/function_calling: Tool calling and agents with streaming support using OCI OpenAI-compatible.
   • langChain/agents: Work in progress – advanced agent workflows and LangGraph examples (check readme).

   Refer to each module's readme for setup, study order, experiments, and Slack channels. Join #igiu-ai-learning for help.

5. OCI commands to manage your files in object buckets (replace NAMESPACE, BUCKET, PREFIX with your values from sandbox.yaml):

   • List objects: oci os object list --all --fields name,timeCreated --namespace NAMESPACE --bucket-name BUCKET --prefix PREFIX
   • Upload file: oci os object put --namespace NAMESPACE --bucket-name BUCKET --prefix PREFIX --file your_file.txt
   • Bulk delete: oci os object bulk-delete --namespace NAMESPACE --bucket-name BUCKET --prefix PREFIX

Create a .env file at the project root for sensitive values referenced in sandbox.yaml (e.g., DB_PASSWORD).

Example .env:

MY_PREFIX=your_oracle_id
DB_PASSWORD=your_db_password

Dependencies like python-dotenv are installed via uv sync. Load in code: from dotenv import load_dotenv; load_dotenv().

• This workshop focuses on OCI GenAI library (https://github.com/oracle-samples/oci-openai) and LangChain 1.0.3.
• Keep code simple for teaching; each file has comments on purpose, docs, env, run command, key sections.
• Update module readmes as you experiment.
• For issues, use #igiu-ai-learning Slack.