Syllabi.ai

Inspiration

Teachers don't struggle because AI can't write lesson plans. They struggle because a good lesson has dozens of interconnected constraints.

For example:

Curriculum standard
       +
Learning objective
       +
Class duration
       +
Student level
       +
Accessibility needs
       +
Teaching modality
       +
Assessment requirements
       ↓
   GOOD LESSON

And when one constraint changes:

The entire lesson may need to change.

That is much more interesting.

Educators routinely spend more time formatting syllabi, constructing grading rubrics, and breaking down multi-week lesson plans than actually teaching—a massive administrative drain that pulls energy away from classroom instruction and student mentorship. We built Syllabi.ai to serve as an intelligent education platform that turns simple instructional notes into beautifully structured, production-ready curriculum materials in seconds. Instead of treating curriculum planning as a one-shot prompt, Syllabi.ai turns it into a guided, constraint-aware collaboration between teacher and AI.

Don't prompt for a lesson. Collaborate with an agent to design one.

What it does

Understands Before It Generates: Syllabi.ai asks targeted questions about class duration, learning objectives, student level, and accommodations before generating content—reducing generic, one-size-fits-all outputs.

Teacher-in-the-Loop Planning: The system proposes a structured lesson outline before generating the final materials. Teachers can review, modify, or approve the structure, keeping the educator in control.

Structured Pedagogical Generation: Transforms simple topic inputs into multi-week syllabi, detailed daily lesson blueprints, and tailored assignments aligned with specific learning objectives.

Targeted Assessment & Quiz Creation: Generates customizable quizzes, answer keys, and grading rubrics instantly—spanning multiple formats like multiple-choice, short answer, and application-based essay prompts.

Constraint-Aware Differentiation:

Crafts interactive lab activities, step-by-step experiment instructions, real-world case studies, and practical project prompts complete with safety guidelines and equipment checklists.
Configures lesson plans based on visual needs, automatically generating diagram descriptions, slide deck outlines, whiteboard layout suggestions, or text-only descriptions when visual aids are toggled off.

Universal Audience Adaptability: Dynamically tailors vocabulary, complexity, and tone across the entire learning spectrum—from engaging K-12 activities to rigorous undergraduate course structures and specialized continuing-education modules for working professionals.

Custom Pace & Modality Tuning: Adapts content structure for accelerated bootcamps, traditional semester-long courses, self-paced modules, or hybrid/remote teaching environments.

How we built it

Gemini -> Google GenAI SDK -> Python/Streamlit -> Docker -> Artifact Registry -> Cloud Run -> Secret Manager

AI Core: Powered by Google's Gemini model gemini-3.6-flash, integrated via the unified, official google-genai Python SDK for fast, high-quality document generation.

Frontend: Interactive user interface developed with Streamlit for rapid, intuitive document generation and real-time output rendering.

Containerization: Packaged as a lightweight, dockerized container and stored securely in Google Cloud container storage (Google Artifact Registry / GCR).

Serverless Hosting: Deployed on Google Cloud Run with auto-scaling configured handling traffic spikes effortlessly.

Firestore: Session history is managed via Firestore, ensuring seamless cross-session context persistence without server maintenance.

Security: Utilized Google Cloud Secret Manager to inject the GEMINI_API_KEY at runtime via principle-of-least-privilege IAM roles (roles/secretmanager.secretAccessor), ensuring zero secrets are exposed in code or container image layers.

Integrated Google Cloud Ecosystem: Engineered as a multi-service pipeline leveraging Google Cloud Run (compute), Artifact Registry/GCR (container management), and Secret Manager (security) working in sync.

Challenges we ran into

IAM & Cloud Security: Debugging initial container-to-service-account permissions to safely grant Cloud Run access to encrypted secrets in Google Cloud Secret Manager (roles/secretmanager.secretAccessor) without exposing API keys in source code or Docker image layers.

API Rate Limits & Quota Exhaustion: Navigating Gemini API free-tier quotas and concurrency constraints—where hit-rate caps restrict active concurrent user usage to a maximum of 2 active chat sessions before running into 429 quota-exceeded limits.

Decoupled Architecture: Eliminating hardcoded AI dependencies by building a dynamic environment configuration layer so underlying Gemini model gemini-3.6-flash can be swapped instantly without rebuilding Docker container images.

Container Startup Optimization: Streamlining Docker container image layers and dependency configurations to minimize startup latency and handle serverless cold starts efficiently on Cloud Run.

Accomplishments that we're proud of

Production-oriented cloud architecture Delivered a fully containerized, secure, serverless cloud pipeline—not just a local prototype.

Teacher-in-the-Loop AI Kept educators in control by requiring approval of the proposed lesson structure before final generation.

Model-Agnostic AI Core Separated model configuration from the application so Gemini model versions can be changed through runtime configuration without rebuilding the application.

Constraint-Aware Curriculum Workflow Built a guided workflow that moves from teacher intent -> clarification -> outline -> approval -> differentiated content instead of relying on a single generation prompt.

Constraint-Aware Replanning When a teacher changes a constraint - such as reducing a 50-minute class to 35 minutes, changing the learner level, or adding an accommodation, Syllabi.ai will identify the affected parts of the curriculum and intelligently re-plan them without starting from scratch.

What we learned

Cloud Security Best Practices: Hands-on experience configuring Google Cloud IAM roles, compute service accounts, and Secret Manager access bindings.

Modern Google GenAI SDK: Implementing the standard google-genai client for clean, robust generation workflows.

Serverless Containerization: Mastering Docker container deployments, Artifact Registry management, and networking on Cloud Run.

What's next for Syllabi.ai

Direct LMS & Document Export: One-click exports to PDF, Google Docs, Canvas, and Google Classroom.

Multi-Agent Orchestration: Upgrading from standard generation calls to autonomous multi-agent workflows using the Google Agent Development Kit (ADK) for dynamic curriculum design.

Multimodal Syllabus Conversion: Allowing educators to snap photos of existing physical textbooks or notes to instantly generate matching quizzes, slides, and lesson plans.

Classroom Ecosystem: Integrate with platforms such as Google Classroom and LMS systems to move from curriculum creation to classroom delivery, assessment, and feedback in one continuous workflow.