T.A.W.S

An AWS-powered AI teaching assistant that uses Socratic guidance to help students learn, strictly refusing to do their homework for them.

Comment

💡 Inspiration

The theme of AWSHacks 2026 is "Build with Gratitude." As college students , we wanted to use this opportunity to give back to the educators and mentors who have shaped our academic and professional journeys. We are deeply grateful for the guidance of mentors and their dedication to teaching and research inspired us to build a tool that supports the very mission they lead every day. Currently, the rise of generative AI has placed professors in a difficult position where they must act as "plagiarism police" rather than mentors. T.A.W.S. is our way of saying thank you—by providing a tool that handles 24/7 student support while fiercely protecting the academic integrity of their courses.

⚙️ What it does

T.A.W.S. is a specialized AI teaching assistant engineered specifically for educational integrity.

  • The Socratic Loop: Instead of giving direct answers, T.A.W.S. breaks problems down into manageable steps and ends responses with open-ended questions to keep the student engaged.
  • Strict Contextual Grounding: If a course policy or deadline isn't covered in the professor's uploaded syllabus, the AI admits it doesn't know rather than hallucinating an answer.
  • Anti-Cheating Guardrails: Plain-English and programmatic rules actively intercept and block the generation of direct homework solutions.

🛠️ How we built it

We utilized a fully serverless, event-driven AWS architecture:

  • Frontend: Built with React/Vite and hosted on AWS Amplify for global, low-latency delivery.
  • Backend Pipeline: Secure REST routing via Amazon API Gateway with strict CORS management, triggering an AWS Lambda (Python) function that manages session handling (Session IDs) for persistent conversation context.
  • AI & Orchestration: Powered by Amazon Bedrock. We utilized the Amazon Nova Lite foundation model for its fast, pedagogical reasoning.
  • RAG (Retrieval-Augmented Generation): Course materials are stored in Amazon S3 and vectorized into Amazon OpenSearch Serverless to strictly ground the AI's answers in actual course data.
  • Security: Bedrock Guardrails serve as the frontline defense against prompt injections and homework delegation.

🛑 Challenges we ran into

Building a complex, serverless AI architecture from scratch taught us a lot the hard way!

  1. The API Routing Trap: We initially struggled to connect our local Vite development server to the AWS cloud, learning how to properly configure proxy targets to avoid CORS and "Missing Authentication Token" errors.
  2. The "Amnesia" Problem: Our AI kept forgetting the context of the conversation after every message. We had to architect a stateful Session ID system between our React frontend and Lambda backend to maintain the Bedrock Agent's short-term memory.
  3. Lambda Timeouts: RAG lookups take time. We experienced 500 Internal Server errors because our Lambda function was hitting its default 3-second timeout before OpenSearch could return the syllabus data. We extended the timeout and optimized the flow to achieve a stable ~6-second response time.
  4. The Frozen Alias Trap: We had to debug Bedrock Agent versioning, learning to point our backend to the active working draft (TSTALIASID) so our updated system prompts would actually take effect.

🏆 Accomplishments that we're proud of

We successfully built a 100% serverless application that securely connects a modern web frontend to advanced AI orchestration without managing a single EC2 instance. We are incredibly proud of getting the Bedrock Guardrails finely tuned—it flawlessly intercepts cheating attempts ("give me the answer to question 1") and pivots to a helpful, encouraging hint without breaking the user experience.

📚 What we learned

We gained deep, hands-on experience with the AWS ecosystem. We learned how to stitch together API Gateway, Lambda, and Bedrock into a cohesive pipeline. More importantly, we learned about "Pedagogical Prompt Engineering"—how to instruct an AI not just to retrieve data, but to act as a patient, empathetic tutor.

🚀 What's next for T.A.W.S.

Our next major milestone is the Professor Telemetry Dashboard. We plan to implement real-time analytics to give instructors insights into which concepts the class is struggling with the most. We also want to add an automated alert system that flags repeated guardrail breaches, allowing for targeted instructor intervention when students continuously try to bypass academic policy.

Built With

Share this project:

Updates