The Recommended Learning Path:
Opiniated tech stack can be swapped with any Language and framework.
-
Foundations: Begin with Python, build APIs using FastAPI, and master SQL databases. AI models are treated as just another component within a standard software architecture.
-
Model Interaction: Learn to integrate Large Language Models (LLMs) by making API calls, monitoring metrics like token usage and latency. Open Router is suggested for accessing multiple models through a single API key.
-
Structured Outputs: Use Pydantic to enforce specific JSON response formats, ensuring predictability for your backend services.
-
Embeddings & Semantic Search: Implement vector search using PGVector in PostgreSQL to find information based on meaning rather than exact keywords.
-
Retrieval Augmented Generation (RAG): Combine search and generation by injecting relevant document context into the model’s prompt.
-
Tools: Provide agents with access to deterministic functions, such as database lookups, to ensure accuracy for critical tasks.
-
Agents: Build autonomous, multi-step systems that observe, decide, and act. This is recommended as a late-stage step after mastering individual components.
-
Memory & Context: Manage persistent state and context windows to ensure agents handle long-lived conversations efficiently.
-
Evals & Testing: Establish observability and systematic logging to evaluate and improve your AI application’s performance.