Mastering OpenAI & Google Gemini APIs: Build AI Applications

Explore how transformers use attention to power generative AI, explaining encoding with n-grams, inference matrices, and prompts, and why non-deterministic outputs lead to confident hallucinations.

Explore the basics of prompt engineering to optimize large language models for tasks like question answering, text summarization, information extraction, and code generation.

Explore zero-shot and few-shot prompting, distinguishing how large language models respond with no examples versus guided examples, including limitations with complex reasoning and numerical tasks.

Master chain-of-thought prompting to improve logical and mathematical reasoning by showing the model's work through few-shot and zero-shot prompts, aided by context and step-by-step thinking.

Learn how to evaluate prompts and models with structured evals, rubrics, and scores. Explore inputs, model responses, and evaluation types—from multiple choice to rubric-based open-ended assessments—to iteratively improve prompts.

Learn a practical framework for writing powerful prompts, covering context, tone, background data, rules, examples, history, and step-by-step thinking, plus testing and edge-case handling for production-ready prompts.

Grounding enhances LMS accuracy by supplying domain-specific data through retrieval augmented generation, embedding, and semantic search, reducing hallucinations.

Learn exemplars as a prompting technique to guide an LM in structuring responses from text. The lecture shows a Python prompt example with input-output formats and notes chat completion deprecation.

Encoders convert prompts into embeddings that decoders use to generate text, weighting context to determine word relevance. Detailed prompts boost specificity, and multimodal embeddings extend to audio and video.

Explore retrieval augmented generation (rag) to enhance AI answers by surfacing internal and external data via embeddings and a vector store, boosting accuracy, trust, and privacy-aware use cases.

Explore the Model Context Protocol (MCP) as a standard linking data sources to AI models via a client-server architecture. Harness tools, resources, prompts, and sampling to enable multi-step AI workflows.

Define AI agents as software that interact with the environment to achieve goals through multi-model workflows, using task generation, prioritization, an execution agent, and a vector memory store.

Compare closed source and open source llms, and weigh self-hosting against managed hosting to understand data control, maintenance, and cost implications.

Create an OpenAI account, open the developer console, and generate a secret API key with a memorable nickname. Copy and securely store the key, and never share it publicly.

Set up your environment by installing Python on Mac, Windows, or Linux, or using Replit in the browser, then install the OpenAI library with pip and verify installation.

Learn how to price and cap OpenAI API usage, compare GPT-4 and GPT-3.5 Turbo token pricing, and configure hard, soft, and approved monthly limits to avoid surprise bills.

Explore a high level overview of the OpenAI API, including chat completion and text completion endpoints, zero-shot prompting, and capabilities like image generation, transcription, file parsing, embeddings, and content moderation.

Explore the ai application architecture for retrieval augmented generation, using a data store, embedder, lane chain embeddings, and a vector database to enable fast semantic search.

Build a retrieval augmented generation pipeline from scratch, using embeddings and a Lance DB vector store for semantic search, to produce factually accurate answers with citations.

Understand the deprecation of the completions API and the shift to chat completions in the OpenAI API. Use GPT 3.5 turbo instruct for completions, as turbo is chat-only.

translate an article using the OpenAI API through zero-shot and few-shot prompting, with a live translation example and guidance on tokens, performance, and cost.

This live example translates an article using the OpenAI API with a Python script, showing secret key setup in environment variables, a translation prompt, and token versus word limits.

Demonstrates summarizing an article with the OpenAI API using zero-shot prompting, placeholders, and token considerations; explains max tokens, word count, and efficiency trade-offs.

Upgrade your OpenAI API project by running the OpenAI migrate tool, which searches Python files and automatically migrates old code to the new API version, updating imports and completion calls.

Demonstrate chat completion with the OpenAI API by building a bot that uses system and user prompts, tracks context across messages, and manages long prompts for performance.

Explore a live demo of chat completion with the OpenAI API, including environment setup, base messages, initial prompts, and few-shot context shaping the chatbot's trajectory.

Explore sentiment analysis with the OpenAI API using few-shot and zero-shot prompting. See how system messages and context shape results and how to implement with the completion API.

Explore sentiment analysis with the OpenAI API, implementing few-shot and zero-shot prompting to classify movie sentiment. Compare chat completion and completion prompts to see how prompts influence results.

Launch your first project by building a chatbot that analyzes earnings transcripts with AI, using the Discount Cash Flows API data and an LM chat API.

Understand the script structure for building ai driven earnings analysis, detailing the main and processing functions, discounting cash flows api calls, and gpt 3.5 turbo content processing.

Run a Python earnings pipeline to perform an LM request on Google data, extract highlights, lowlights, and sentiment, and explore system prompts and logging.

Learn to replace the OpenAI API call with the Google Gemini API, set up Gemini Studio and an API key, and swap the call logic in your app.

Learn to generate and edit images with the OpenAI image generation API using DALL-E 3 and DALL-E 2, including prompts, sizes, edits with masks, and variations.

Generate phone wallpapers using the OpenAI image generation API by comparing Dall-E three and Dall-E two, and exploring prompts, resolutions, hd and standard quality, plus code examples.

Use OpenAI's text to speech API to narrate text in multiple languages and stream real time audio. See a Python example that outputs mp3, flac, or opus with various voices.

Harness Google Vertex AI Studio to generate rich media—images, video, audio, and music—with Vo and Imagine models. Explore text-to-speech, multilingual support, and API enablement for seamless content creation.

Learn to use the OpenAI text to speech API to transcribe a blog post into a speech audio file in Python, with model, voice options, and output handling.

Use the GPT-4 vision API within the OpenAI framework to analyze images, including multiple images, image URLs or base64 data, with fidelity settings and cost awareness.

Demonstrates a calorie counter using the GPT four vision API, accepting images as file paths, URLs, or base64, and returning estimated calories with best estimate prompts.

Explore Gemini API in Google AI Studio, create an API key, and experiment with chat, free form, and structured prompts for multimodal inputs.

Learn to perform Gemini chat completions in google colab by configuring api keys, loading secrets from Secrets Manager, and tweaking generation settings with streaming options and prompt engineering.

Explore how large context windows let a language model reference more of your data for hyper-specific responses. Apply this to code bases, videos, and financial data in multimodal contexts.

Create a Google Cloud account and enable the translate API to translate subtitles across languages. Set up a service account, download the JSON key, and configure credentials for API access.

Explore basic and advanced translation with the Cloud translation API, including simple and bulk batch calls, language detection, and custom models via AutoML for production-ready translations.

Explore a step-by-step Python walkthrough that translates subtitles using the Google Translate API, Webvtt integration, and language pairs to generate translated caption files.

Learn how to translate subtitles to multiple languages using AI-powered translation with the Google Translate API, preserving timestamps, running translate_subtitles.py, and exploring multilingual outputs.

Fine tuning adapts pre-trained models on a small labeled dataset to tailor ai applications with the OpenAI API, enabling customer support, translation, and content generation.

Learn to fine-tune a GPT-3.5 Turbo model with the OpenAI API by preparing and uploading a json dataset, starting a fine-tuning job, and using the resulting model in chat completions.

Use the GPT builder to create custom GPTs for ChatGPT, upload data and instructions, configure conversation starters, and share them with friends or your organization.

Deploy a Python backend app using the OpenAI API with Django or Flask, using environment variables for secret keys and hosting on GCP App Engine with autoscale.

Explore the do's and don'ts of using generative AI, from content creation and prototyping to upscaling and colorization, while avoiding misinformation, plagiarism, and copyright pitfalls.