Generative AI for Startups: Anthropic Claude 3.5 Python API

Learn how to use the messages API to interact with Claude 3.5 models, retrieve API keys, and configure system messages across variants like sonnet, opus, and haiku.

Dive into the Claude 3.5 messages API, detailing message structures, roles, content formats for single and multi-turn conversations, and required parameters like model and max tokens.

Learn how the max token parameter sets the absolute limit for generated tokens in Claude models (cap of 4096), affecting cost, length, and how usage shows input and output tokens.

Apply custom stop sequences in Claude 3.5 to halt output at predefined text, and observe stop reasons in a Sydney top places example.

Discover how the temperature parameter shapes output variety in the Claude 3.5 Python API, from deterministic zero to creative one, featuring experiments with values zero, 0.7, and 1.

Explore top-p (nuclear) sampling in Claude 3.5 responses, showing how cumulative probability determines token selection, and how temperature and code experiments shape output variety.

Explore top-k sampling and its role in controlling next-token choices in large language models, along with temperature and top-b, with Claude 3.5 experiments showing higher top-k yields more varied outputs.

Explore streaming and non-streaming data delivery. Learn how streaming delivers data in chunks for faster initial responses, real-time processing, and memory efficiency, with applications for chat bots and storytelling apps.

Enable streaming by setting the stream parameter to true and using client.messages.create with a model, max tokens, and temperature, while handling events like message start and content block delta.

Learn to implement streaming responses using the Claude 3.5 API in Python, comparing non-streaming and streaming modes and handling message streaming events.

Learn to send images with text in Claude 3.5 messages API by encoding images as base64, supporting jpeg, png, gif, webp, up to 20 images, and adding captions.

Learn to use Claude 3.5 to analyze and describe an image by converting the image to base64, sending it via the Anthropic API, and obtaining a detailed model-generated description.

Generate code from an image using Claude 3.5 to create a python matplotlib bar plot of cloud model metrics (opus, sonnet, haiku), with x as metrics and y as values.

Generate recipes from food images using Claude 3.5, leveraging base64 image data, system prompts, and Anthropic models to produce tailored recipes, including French outputs.

Learn to handle and compare multiple images in Claude 3.5 API by uploading base64 images of a blue suit with brown and black shoes to receive a fashion color recommendation.

Learn to produce json output from Claude 3.5 by using string parsing to extract json, prefill assistant response, stop sequences, wrap json in xml-like tags, and parse with Python's json.loadstring.

master prefilling json structures for Claude 3.5 responses with the Anthropic Python API, turning unstructured text into a structured json object containing name, age, and profession.

Create a json book catalog using Claude 3.5's json capabilities, including title, author with first and last name, publication year, journals, and a ratings object for at least three books.

Learn how Claude 3.5 integrates tools to access external data, extend capabilities, and deliver more accurate, up-to-date responses by chaining actions.

Design and implement json schemas for Claude 3.5 tools, defining inputs, required fields, and tool calls to fetch weather by location using a get weather tool.

Explore methods to enforce specific tool usage in Claude 3.5, including auto, any, and forced tool selection, with system prompts and examples like capital of France.

Build a stock price prediction app using live news, stock price history converted to images, and tools with messages API, then let the cloud 3.5 model analyze and provide recommendations.

Fetch historical stock data using Yahoo Finance in Python, retrieving the last 90 days for a ticker such as MSFT.

Integrate real-time stock data with live news from the DuckDuckGo search library (DGS.news), analyze sentiment to inform stock price predictions, and plot history to feed Claude 3.5 for recommendations.

Convert stock price history from a data frame into a plot image of the last 90 days and encode it as base64 for model input, enabling tool prompts and recommendations.

Craft tools as JSON schemas for three functions—get stock price, get news summary, and generate stock graph—and implement a process tool call to route inputs to the correct function.

Leverage Claude 3.5 to perform stock analysis from a code, generating an an 80-day price graph, current price, news, and a technical and fundamental outlook with moving averages, RSI, and MACD.