LifeOS: Agentic Physics Workspace

LifeOS: Agentic Physics Workspace 🚀

LifeOS is an interactive AI workspace where autonomous agents powered by Amazon Nova collaborate to complete real-world tasks. Instead of interacting with AI through a traditional chat interface, LifeOS lets users watch the reasoning process unfold inside a physics-driven environment where agents, tasks, and opportunities move and interact dynamically.

The inspiration behind LifeOS came from noticing that most AI systems today only answer questions, but they rarely execute workflows. Many everyday activities—like applying for internships—require repetitive steps such as searching for opportunities, tailoring resumes, and submitting applications. We wanted to explore what happens when AI moves beyond conversation and becomes a system that acts on goals.

In LifeOS, the user simply enters a goal such as:

Apply to 3 AI internships in Pune

Once the goal is submitted, a group of specialized AI agents begins collaborating to complete the task. A Planner Agent, powered by Amazon Nova 2 Lite, interprets the user’s goal and converts it into a sequence of actionable steps. A Search Agent then discovers relevant opportunities using semantic embeddings. A Resume Agent generates a tailored resume based on the job description, and finally an Apply Agent completes the workflow by submitting the application.

What makes LifeOS unique is that the entire process is visualized in a physics-based workspace. Using the Matter.js physics engine, tasks appear as floating cards and agents interact with them dynamically—pulling them closer, processing them, and eventually sending completed tasks into a completion zone. Instead of hiding the AI reasoning process behind an interface, LifeOS allows users to see how AI agents think and collaborate in real time.

The system was built using Vanilla JavaScript and Matter.js on the frontend and FastAPI with Python on the backend. AI reasoning and generation are powered by Amazon Nova foundation models via AWS Bedrock, while embeddings are used to match job descriptions with user intent. Semantic similarity between opportunities and the user goal is computed using cosine similarity:

$$ similarity(A,B) = \frac{A \cdot B}{||A|| \cdot ||B||} $$

One of the biggest challenges during development was synchronizing real-time AI reasoning with the physics simulation. The system needed to dynamically create tasks, trigger agent actions, and update the UI without interrupting the physics environment. Building a smooth interaction between AI decisions and visual feedback required careful orchestration between the backend APIs and the frontend simulation.

Through this project we learned how to design agentic AI systems, how embeddings enable intelligent semantic search, and how visualization can make AI reasoning easier to understand. More importantly, we realized that when users can see the reasoning process rather than just the output, AI systems become far more transparent and engaging.

LifeOS explores a future where AI systems behave less like chat tools and more like autonomous digital collaborators. By combining Amazon Nova models, multi-agent workflows, and an interactive physics interface, LifeOS demonstrates a new way to interact with intelligent systems—one where users can literally watch AI turn goals into actions.

Built With

• amazon-nova
• amazon-titan-embeddings
• aws-bedrock
• css
• fastapi
• html
• javascript
• matter-js
• python
• vite