Model Mondays is a weekly series that helps you build your AI Model IQ with 5-min news recaps, 15-min tech spotlights and 30-min AMA sessions with subject matter experts. Join us!
Spotlight On: SLMs and Reasoning
🧪 | This section was generated with AI help and human revision & review. To learn more, please refer to this document in our website.
How can you bring advanced reasoning to resource-constrained devices? This session explores the latest in Small Language Models (SLMs) like Phi-4, which are redefining what’s possible for agentic apps.
Mojan Javaheripi, a Senior Microsoft Researcher, will discuss how SLMs leverage inference-time scaling and chain-of-thought reasoning to deliver powerful results on smaller hardware. Discover use cases, deployment strategies, and how SLMs are making AI more accessible and efficient for everyone.
Download The Deck With Key Resource Links:
Phi-4 Reasoning: Power Through Precision
In her spotlight talk, Mojan introduced the Phi-4-reasoning model and a Phi-4-reasoning-plus model and walked us through the techniques used to create these variants in a manner that allows them to work in resource-constrained environments, while providing comparable accuracy to larger language models. The motivation for this work comes from wanting to bridge the gap between large frontier models and smaller, more efficient models that can run on commodity hardware including personal devices like laptops, but also edge devices at industry scale.
Watch The Replay From Her Livestream
Why is this important?
The Phi-4 reasoning models are designed for explicit, step-by-step problem solving, with a focus on explainability and logical decomposition. Mojan details the training pipeline, including data curation, supervised fine-tuning, and reinforcement learning, and shares benchmarking results that show Phi-4 models outperforming much larger models in math, science, and coding tasks. The discussion covers community adoption, quantization for edge devices, and the importance of fine-tuning for domain-specific applications. Real-world use cases include intelligent tutoring, autonomous agents, code generation, and strategic planning.
Key Takeaways From The Session
- Phi-4 reasoning models deliver state-of-the-art reasoning and efficiency at a fraction of the size of frontier models.
- SLMs are ideal for edge, local, and resource-constrained deployments, with strong support for quantization and customization.
- Fine-tuning & reinforcement learning enable adaptation to new domains & tasks.
- Open-source and community contributions accelerate innovation and adoption.
- Responsible AI, safety, and explainability are core to Phi-4 model design.
Sidebar: AI & Content Generation
You may have noticed the 🧪 annotation about AI-generated content. Let's talk about how this was done, and why.
Model Mondays is a series about building your AI model IQ - and learning how to find the right model for the task. And the best way to learn something is by applying it to a real problem. Starting this week, I'm using AI to streamline many content-related tasks for Model Mondays. And I wanted to share not just the reasoning for this, but the process and tooling I use - in the hopes that it helps others find productive uses for the same workflows.
Motivation: Efficiency
On a weekly basis I have to plan for two different sessions (livestream and AMA), coordinate across speakers and hosts, then publish pre- and post- event content writeups that provide promotion and recap value respectively.
Writing these manually is not just time-consuming, it is also complicated by the fact that each week has new topics to research - and new discussions to summarize that require cognitive bandwidth. My a-ha moment was in recognizing that prompt creation is also content creation and crafting a well-written prompt to create content can be as rewarding and effective as authoring the content from scratch.
Workflow: Agent Mode + MCP
How am I doing this? I'm using Visual Studio Code Agent Mode in a GitHub Codespaces environment launched on the Model Mondays repo. Every task is reduced to a custom prompt that is stored in the codebase for reuse.
Here are some examples:
- Give it the YouTube transcript and have it summarize it
- Give it the AMA transcript and have it generate a recap
- Detect key terms and use MCP Servers to find resource links.
You can simplify this further if you use MCP Servers to extend Agent Mode as shown - adding new "tools" to help the agent perform key tasks.
In my case, there were 3 tools that were immediately helpful:
- Microsoft Docs MCP Server - for related resources
- Hugging Face MCP Server - for Model Hub links
- Azure AI Foundry MCP Server - for Azure AI Foundry Models catalog
Want to see these in action? Check out the "Related Resources" in this post, populated by using the three MCP servers above.
Keep reading this series for more updates and learnings from our AI experiments!
Read The Recap
Want an easy way to catch up on all the news? Check out complete blog series written by our resident student blogger here on Tech Community!
You can also catch her recap for this week, right here on dev.to:
Build Your Model IQ!
Model Mondays Season 2 is currently scheduled to go from June to September, covering the 12 key topics shown below.
- 👉🏽👉🏽 Register for Upcoming Livestreams to get reminders
- 👉🏽👉🏽 Register for Upcoming AMAs to get reminders
This post really "models" the way forward for SLMs! Loving all the reasoning — it's clearly small but mighty brainpower at work.