Generative AI and RAG Development

Store and load API credentials with environment variables without committing them to Git.

Interpret the method, URL, headers, body, status code, and JSON content of an HTTP exchange.

Call a JSON-based web API from Python and extract fields from its response.

Handle API timeouts, rate limits, and invalid responses with bounded retries and clear error messages.

Write and run focused automated tests for a Python function that processes API data.

Explain how tokens, next-token prediction, context windows, and inference affect a language model's output.

Distinguish open-source software, open-weight models, hosted model APIs, and model providers.

Use a model card and license to assess a model's capabilities, restrictions, resource needs, and context length.

Run an instruction-tuned open-weight model locally through a model runner.

Send a prompt to an open-weight model through a provider's free hosted endpoint.

Switch a small Python application between local and hosted models through a common model interface.

Compose system and user messages that specify a task, context, constraints, and examples.

Adjust generation parameters and explain their observed effects on consistency and variation.

Prompt a model to return data conforming to a defined JSON structure.

Validate model-generated structured data and reject or repair invalid output.

Create a small evaluation set and use explicit criteria to compare two prompts or models.

Diagram the ingestion, retrieval, and generation stages of a RAG system and identify when RAG is appropriate.

Extract text from source documents and divide it into chunks with traceable metadata.

Generate embeddings for document chunks and store them in a local vector index.

Retrieve and rank document chunks for a user's question.

Generate an answer that cites retrieved sources and abstains when the sources are insufficient.

Evaluate retrieval and answer quality with a prepared question set and diagnose a failed example.

Distinguish a deterministic workflow from an agentic workflow and select between them for a given automation task.

Define a typed agent tool with a narrow purpose, validated inputs, and deterministic outputs.

Build a stateful workflow graph with nodes, transitions, and shared state using an open-source orchestration framework.

Route model decisions to approved tools while enforcing iteration and timeout limits.

Add a human approval step before an agent performs a consequential or external action.

Save workflow checkpoints so an interrupted run can be inspected and resumed.

Add execution logs and graceful handling for model, retrieval, and tool failures.

Explain the context-gathering, edit, test, and revision loop used by coding agents.

Configure an open-source coding assistant to use an available local model or free hosted model.

Write a bounded coding task specification and provide only the relevant repository context.

Use a coding agent to implement a small, test-backed feature in a Git repository.

Review an agent-generated code diff, run its tests, and accept or revert the change.

Define user stories, permission boundaries, and acceptance tests for a document-to-action workflow assistant.

Design the app's interface and the boundaries among its model, retrieval, agent, tool, and storage components.

Build an interface that accepts documents and a workflow request and displays progress and results.

Connect document ingestion and cited retrieval to the app's workflow.

Implement tools that extract action items, draft a response, and export an approved result.

Orchestrate retrieval, tool use, and human approval through a bounded agent workflow.

Run acceptance, failure, and safety tests and correct the highest-impact problems.

Package the app with reproducible setup instructions for local Windows use and a browser-based alternative.