FastHTML - Web Development with Pure Python + Hamel's open source course | High Signal AI #11
Explore FastHTML's Python-powered web development, LangGraph Studio's visual AI workflows, Hamel's open LLM course, generative AI platform building guide, Merlinn's AI on-call developer, and more!
We are about to reach 1K subscribers. New subscribers should note that the purpose of this newsletter is to help you build with AI and keep you focused on building, that’s why the updates / news section is not prioritised. Only important updates are from models and libraries.
"Don't find customers for your products, find products for your customers." - Seth Godin
Highlight of the week 🏆: FastHTML
GitHub repositories you must check out
AI startups that raised funds
High value learning resources
Reading list for the week
What people are building with AI
Updates from dev libraries, hubs and tools
Highlight of the week 🏆: FastHTML
I have only tried FastHTML for a handful of applications and I can say that this is definitely the first framework that enables you to develop modern web applications in pure python.
Streamlit is restricted to just a bunch of data analysis use cases and that too for prototypes or internal usage but with FastHTML you can actually add JS, CSS and other components as in a web application.
In fact, the home page of FastHTML is actually a FastHTML app.
What type of applications can you build with it?
It's good for: general purpose web applications (i.e anything you'd build with React, Django, NexJS, etc); quick dashboards, prototypes, and in-company apps (e.g. like what you might use gradio/streamlit/etc for);
Analytics/models/dashboards interactive reports; Custom blogs and content-heavy sites where you also want some interactive/dynamic content.
Where can you deploy FastHTML to?
You can deploy a FastHTML app to any service or server that supports Python. They have guides and helpers for Railway.app, Vercel, Hugging Face Spaces, Replit, and PythonAnywhere.
Find out more on their webpage.
GitHub repositories you must check out
LangGraph Studio by LangChain AI (138 stars) - LangChain released LangGraph Studio (Beta) - A visual tool for creating and managing complex workflows with LangChain, allowing users to design and test AI-driven applications with ease.
Visual Workflow Design: LangGraph Studio provides an intuitive graphical interface for designing and managing workflows, which helps streamline the process of integrating and experimenting with LangChain's features.
Modular Components: The tool supports a modular approach, allowing users to easily add, configure, and connect different components and services within their workflows.
Real-Time Testing and Debugging: It offers real-time testing and debugging capabilities, enabling users to monitor and troubleshoot their workflows directly within the tool.
Integration with LangChain: The repository is tightly integrated with LangChain, making it easier for users to leverage the full capabilities of the LangChain framework in their projects.
Guide to make your own SWE Agent -
swekitis a framework for building SWE agents on by utilising composio tooling ecosystem. This directory contains Python code related to software engineering (SWE) within the Composio project. It focuses on implementing and managing components for SWE tasks.SWE Kit allows you to
Scaffold agents which works out-of-the-box with choice of your agentic framework,
crewai,llamaindex, etc...Tools to add or optimise your agent's abilities
Benchmark your agents against
SWE-benchThepython/swedirectory contains
AI startups / businesses that raised funds
Crypto-AI startup Mira raised $9 million in a seed round co-led by Bitkraft Ventures and Framework Ventures.
Protect AI, a Seattle-based cybersecurity startup, raised $60 million in a funding round led by Evolution Equity Partners.
OpenAI-backed healthcare AI startup Ambience Healthcare raised $70 million in a Series B funding round.
High Value Learning resources 📚
An Open Course on LLMs, Led by Practitioners - This would be the 2nd best find for this week for me. Hamel is a gem of an instructor and he, along with other practitioners have released a set of workshops and talks on topics like evals, retrieval-augmented-generation (RAG), fine-tuning and more.
Course Overview: The blog post provides an overview of a course designed to teach modern software engineering practices, focusing on key principles such as testing, continuous integration, and deployment.
Learning Objectives: The course aims to equip learners with practical skills for developing high-quality software, including writing robust code, implementing automated testing, and using tools for effective project management.
Target Audience: It is targeted at developers looking to improve their software engineering skills and gain hands-on experience with current best practices and tools used in the industry.
[Course] Embedding Models: From Architecture to Implementation
Comprehensive Understanding of Embeddings: The course provides a detailed exploration of embedding models, covering their architecture, functionality, and how they are applied in various machine learning tasks.
Hands-On Implementation: You will gain practical experience by implementing embedding models, allowing them to understand not only the theoretical aspects but also the practical challenges and solutions in real-world applications.
Advanced Techniques and Best Practices: The course emphasizes advanced techniques and best practices in designing and deploying embedding models, equipping learners with the skills needed to handle complex data and achieve effective model performance.
Reading list for the week
Building A Generative AI Platform - The post provides a detailed walkthrough of building a generative AI platform, starting from a basic setup and progressively adding components to address growing complexities. You will gain a thorough understanding of the key components needed for a production-ready AI system.
Practical insights on system design tradeoffs: The article discusses important considerations and tradeoffs for each component, such as latency vs. reliability for guardrails, or the pros and cons of different caching techniques.
Focus on reliability, security and performance: The post emphasizes critical aspects like guardrails, observability, and caching that are essential for building robust, secure and efficient AI systems at scale. Readers will learn best practices for mitigating risks and optimizing performance in real-world AI deployments.
What people are building with AI 🧑💻
merlinn by merlinn-co - Open source AI on-call developer 🧙♂️ Get relevant context & root cause analysis in seconds about production incidents and make on-call engineers 10x better.
Merlinn is an AI-powered on-call engineer agent that lives in Slack and helps troubleshoot production incidents by connecting to your favorite tools and pair with you to triage the incident.
It uses langchain behind the scenes for creating the agent and connecting it to tools like DataDog, PagerDuty, Opsgenie, etc.
LangGraph-multi-agent-user-facing notebook from the ai-champ-design-patterns - The notebook demonstrates how to design and implement a multi-agent system using LangGraph, focusing on creating user-facing applications that leverage multiple AI agents to interact with users effectively.
Updates from dev libraries, hubs and tool
Llama 3.1 Breakthrough: Meta's Llama-3.1-405B reached #3 on the Overall Arena leaderboard, becoming the first open model in the top 3. It excels in coding, math, and following instructions.
SAM 2 Launch: Meta unveiled Segment Anything Model 2, enhancing video and image segmentation. It processes 44 frames per second, needs 3x fewer interactions, and speeds up video annotation by 8.4x compared to manual methods.
OpenAI Voice Rollout: OpenAI is testing an improved Voice Mode with select Plus users, aiming for a fall rollout to all Plus subscribers. The feature promises more natural real-time conversations.
Perplexity AI’s Publishers Program: Perplexity AI introduced a Publishers Program featuring TIME, Der Spiegel, and Fortune. They also added a status page for their product and API.
LlamaIndex Adopts MLflow: LlamaIndex integrates MLflow to streamline model lifecycle management, from development to deployment.
Nvidia’s Project GR00T: NVIDIA's Project GR00T offers a structured approach to scaling robot data. It includes collecting human demonstrations with Apple Vision Pro, amplifying the data using RoboCasa (a generative simulation framework), and enhancing it further with MimicGen.
Prompt Tuner: Cohere has introduced Prompt Tuner in beta, a tool that allows users to refine prompts directly within their Dashboard through a customizable optimization and evaluation loop.