Qwen 32B Beats DeepSeek R1

Today’s top AI Highlights:

1. Self-improving AI agents that work directly with your tools and data

2. Microsoft’s opensource SDK to build AI agents in C#, Python, and Java

3. Alibaba Qwen’s new 32B model performs at par with DeepSeek R1 671B

4. Windsurf agent can now see and fix UI issues

5. OpenAI plans to charge $20,000 per month for specialized AI agents

& so much more!

Read time: 3 mins

AI Tutorials

AI Agent Tutorial

Air quality has become a crucial health factor, especially in urban areas where pollution levels can significantly impact our daily lives. While many air quality monitoring tools exist, there's a gap when it comes to personalized health recommendations based on real-time air quality data.

In this tutorial, we'll walk you through building a multi-agent AQI Analysis App that gives personalized health recommendations based on real-time air quality data. This system will analyze current air conditions and provide tailored advice based on your health conditions and planned activities.

Tech stack:

• Firecrawl for web scraping

• Agno (formerly Phidata) to create and coordinate AI agents

• OpenAI GPT-4o as LLM

• Streamlit for interface

AI Workflow

This workflow combines Grok-3's image generation capabilities with Pika AI's video animation features to create stunning transformation videos that show the evolution from vintage to modern aesthetics. Perfect for photo restorations, concept visualizations, or creative storytelling.

We share hands-on tutorials like this 2-3 times a week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads, Facebook) to support us!

Latest Developments

First Autonomous AI Agent Platform Built for MCP ⛓️🧰

Latitude, an open-source platform for prompt engineering, has launched its autonomous agent platform built around the Model Context Protocol (MCP). This platform lets you build, test, and deploy self-improving AI agents that can directly interact with your existing tools and data sources.

Latitude handles the complex task of hosting and securing MCP servers, so you can focus on agent logic and integration. Starting with support for over 20 MCP servers, they expect to reach 100+ by month's end. They are also offering a free tier that includes up to 40,000 prompt and evaluation runs per month.

Key Highlights:

1. Creating Autonomous Agents - Built specifically for autonomous agents rather than predefined workflows, Latitude enables agents to plan and operate independently with just initial instructions and a set of tools. The platform includes an agentic runtime that loops until goals are reached.

2. Integration via MCP - Connect your agents to various data sources (databases, APIs, files) and tools using the MCP. This eliminates the need for custom integrations for each connection, providing a secure and consistent interface.

3. Prompt Engineering and Evaluation - Latitude includes a collaborative prompt manager with version control, a playground for testing, and robust evaluation tools (LLM-as-judge, code-based, human feedback). This allows for iterative improvement and ensures prompt quality.

4. Production-Ready - Latitude hosts MCP servers in isolated Docker containers within a private VPC, handling authentication and network security. This enables a smooth transition from development to deployment, with built-in observability and logging.

Semantic Kernel to Build AI Apps in C#, Python, and Java 📝🔌🤖

Semantic Kernel by Microsoft is an open-source SDK to streamline the process of creating AI agents and integrating cutting-edge AI models into your existing projects. This toolkit works seamlessly with C#, Python, and Java, providing a middleware layer that handles the complex communication between your code and LLMs.

One of the strongest features is that it also allows building agentic apps where multiple agents can collaborate within a single conversation, and even with human inputs. And, this lightweight SDK is specifically built to be future-proof, so you can easily swap in newer, more powerful AI models as they emerge, without a complete code overhaul.

Key Highlights:

1. Centralized Kernel Architecture - The Kernel acts as the core, managing services, plugins, and the entire AI execution flow. This centralized design simplifies configuration, monitoring, and implementation of responsible AI practices.

2. Function Calling with Plugins - Semantic Kernel makes it easy to integrate existing code (APIs, services) as "plugins." You expose functions with semantic descriptions, and the AI intelligently calls them based on user requests, handling the complex orchestration. Supports native code and OpenAPI.

3. Orchestration with Planners - SK includes "planners" that take high-level goals. The AI automatically creates and executes a plan to achieve these, calling the necessary plugin functions. This significantly reduces the code you have to create for complex AI interactions.

4. RAG Support - SK has built-in capabilities for Retrieval Augmented Generation (RAG). It seamlessly connects to and manages various vector databases (Azure AI Search, Qdrant, Chroma, etc.), enabling powerful knowledge-enhanced AI.

5. Enterprise-Grade Observability - Designed for production, SK emits detailed logs, metrics, and traces compatible with OpenTelemetry. You can effectively monitor your AI agents, functions, and overall performance.

Quick Bites

Google has expanded its OpenAI compatibility layer for the Gemini API, giving you access to more Gemini models and capabilities through the OpenAI SDK, with just 3 simple code changes. This update enables audio inputs, image generation, function calling, and structured JSON outputs while maintaining compatibility with OpenAI's Python and JavaScript libraries.

Google has launched an experimental AI Mode, powered by a custom version of Gemini 2.0, that expands AI Overviews with enhanced reasoning and multimodal capabilities for complex, multi-part questions. The new feature issues multiple related searches simultaneously across various data sources to deliver comprehensive responses with web links, making it particularly useful for nuanced comparisons and in-depth explorations. Currently available to Google One AI Premium subscribers.

LangChain has launched langgraph-bigtool, a Python library for LangGraph AI agents to handle vast numbers of tools, boosting their capabilities significantly. The library allows agents to access and manage extensive toolsets, with options for custom retrieval functions and metadata storage using LangGraph's persistence layer. You can install it via pip install langgraph-bigtool.

xAI’s latest Grok 3 now tops the Lmsys Chatbot Arena leaderboard! With over 3k votes, grok-3-preview-02-24 is tied for #1 overall, and across Hard Prompts, Coding, Math, Creative Writing, Instruction Following, and Longer Query, outperforming GPT-4.5 and Gemini 2.0 models.

OpenAI has started rolling GPT-4.5 to all ChatGPT Plus subscribers with a “sizable rate limit,” which will change based on the demand for the model. The rollout will be completed in the next 1-3 days.

Here’s a big one from OpenAI though. Reportedly, OpenAI is planning to launch of series of AI agent products, tailored for different applications, including sorting and ranking sales leads and software engineering. For instance, a “high-income knowledge worker” agent, will reportedly be priced at $2,000 a month. Another software developer agent is said to cost $10,000 a month.

Alibaba’s Qwen team has released QwQ-32B, a model with 32 billion parameters that achieves performance comparable to DeepSeek-R1, which boasts 671 billion parameters (with 37 billion activated). This efficiency is due to scaled reinforcement learning, enhancing areas like math and coding, and using outcome-based rewards. Furthermore, the team has integrated agent-related capabilities into the reasoning model to think critically while utilizing tools and adapting its reasoning based on environmental feedback. You can download QwQ-32B via Hugging Face and Modelscope.

Windsurf has released Wave 4, a major update to its agentic IDE that introduces several groundbreaking features for smoother coding workflows.

• Previews: You can now click directly on UI components you want to modify or automatically capture error logs from your running applications that automatically feed back to Cascade as context

• Tab-to-import: It now detects when you're using new dependencies and offers to automatically add the required import statements at the top of your files with a simple tab press

• Linter Integration: Cascade now automatically checks its generated code against linters and fixes any style or syntax issues before presenting it

• Suggested Actions: After responding to a query, Cascade now recommends logical next steps to keep you in a productive flow state

Tools of the Trade

1. Defog AI's Introspect: Open-source tool for deep research across structured data (databases, CSVs), unstructured data (PDFs), and the web. It uses Claude-powered agent with recursive tool use (SQL generation, web search, PDF analysis) to answer user questions.

2. Groovy: Python-to-JavaScript transpiler, meaning that it converts Python functions to their JavaScript equivalents. It is used in the Gradio library, so that developers can write functions in Python and have them run as fast as client-side JavaScript.

3. HelixDB: Lightweight high-performance multi-model graph database built in Rust that offers native vector support, lightning-fast startup times and millisecond query latency, making it ideal for AI applications and RAG implementations.

4. Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.

Hot Takes

1. For 25% of the [YC] Winter 2025 batch, 95% of lines of code are LLM generated.
   That’s not a typo. The age of vibe coding is here. ~
   Garry Tan
   
   

2. People in the X AI community are 3 years ahead of everyone else, not even joking.
   We're so far ahead that for 90% of people, AI just means ChatGPT. ~
   Ashutosh Shrivastava

That’s all for today! See you tomorrow with more such AI-filled content.

Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!

Unwind AI - X | LinkedIn | Threads | Facebook

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉