AWS Builder Center
Ctrl+F is Dead: How I Built an AI Semantic Engine for College Courseware with AWS Serverless & Bedrock

Ctrl+F is Dead: How I Built an AI Semantic Engine for College Courseware with AWS Serverless & Bedrock

Ever tried finding a specific networking slide at 3 AM before an exam, only to dig through five files named "Unit4_draft_final.pptx"? NotesForAll AI fixes this. Built with Angular and AWS Serverless (Bedrock, Lambda, DynamoDB, and S3), it transforms messy courseware into an AI engine that semantically searches lecture decks in milliseconds, auto-generates exam FAQs with Amazon Titan, and streams slides with zero lag.

The 3 AM Pre-Exam Ritual Nobody Talks About

It is 3:15 AM. You have an engineering exam at 9:00 AM.
You distinctly remember the professor showing a crucial diagram explaining how "Transport Layer socket connections and multiplexing" work. You open your class repository. What greets you?
  •  presentation_final.pptx 
  •  Unit_IV_Updated_v2.pdf 
  •  Notes_Prof_Copy(1).docx 
  •  Scan_2026_09_12_053518.pdf 
You hit  Ctrl+F  and search "transport layer socket connections".
Zero results. Why? Because  Ctrl+F  is illiterate. It only matches exact string characters inside open tabs, completely ignoring slides formatted with custom shapes, converted PDFs, or documents whose filenames look like MD5 hashes.
I decided enough was enough. Students don't need another generic Google Drive folder; they need an intelligent, semantic-aware courseware platform that understands what's inside the slides and does the studying legwork for them.
Meet NotesForAll AI.



What NotesForAll AI Actually Does

NotesForAll AI is an open academic repository and AI study engine. Instead of forcing you to hunt through dozens of PDFs and PowerPoint decks, it brings modern Generative AI and vector search directly into the lecture hall.
Here are the superpowers it ships with:

1. Semantic Search That Reads Your Mind (Not Just Keywords)

If you search for "How does a connection-oriented 3-way handshake work?", traditional portals return nothing unless the file is literally titled  3-way-handshake.pdf .
NotesForAll AI uses Amazon Bedrock (Amazon Titan Embeddings). During document ingestion, multi-page slides and PDFs are broken into semantic chunks, vectorized into 1,536-dimensional embedding vectors, and indexed in DynamoDB. When you type a query:
  • Your query is converted into a vector representation in real time.
  • A hybrid vector cosine similarity + lexical ranking algorithm scores every chunk.
  • Within under 200ms, you get the exact slide, page number, and paragraph snippet, even if your search terms didn't match the exact words on the slide.

2. Auto-Generated Exam Cheat Sheets & Study FAQs

Finding the right slide is half the battle; knowing what the professor might ask in the exam is the other.
Upon ingestion, Amazon Titan Text analyzes the document's contents and synthesizes:
  • A Modular Concept Summary: A 2-minute executive brief of the material.
  • High-Yield Exam FAQs: 3 to 5 examination-style questions with model answers (e.g., "What is the difference between connection-oriented vs connectionless services?").
No more prompting random external chatbots and hallucinating syllabus topics. The FAQs live directly inside the document viewer, side-by-side with the original text.

3. Zero-Wait Document Streaming

College slide decks are notoriously huge. PPTXs and scanned PDFs regularly clock in at 10MB to 50MB. If your backend tries to buffer that into Lambda memory before serving it to the browser:
  1. Your memory bill goes through the roof.
  1. Lambda hits its 6 MB response payload limit and dies.
  1. You stare at a spinning loader until you lose the will to study.
Instead, NotesForAll AI uses an S3 Presigned Redirect Architecture. The API Gateway router issues a lightweight  HTTP 302  redirect to an AWS S3 presigned URL. The browser then uses native  HTTP 206 Partial Content  byte-range streaming to view 50-slide decks smoothly, with no file bytes ever passing through Lambda.

4. Distraction-Free Neo-Brutalist UI

Modern web design is often cluttered with bloated animations and cookie banners. NotesForAll AI is built with Angular 17+ using a custom Neo-Brutalist Bento design system:
  • High-contrast typography and bold retro borders.
  • Instant keyboard navigation with inline search modals.
  • Interactive slide table of contents (TOC) with scrollspy navigation.



The Serverless Engine Under the Hood

Here is how the request flow works behind the scenes on AWS:
[ Student / Browser ] (Angular 17 on AWS Amplify)
|
|-- 1. Semantic Search Query (e.g., "socket connections")
| |
| v
| [ Amazon API Gateway ]
| |
| v
| [ AWS Lambda (Node.js/TypeScript) ]
| |
| |-- [ Amazon Bedrock ] (generate Titan vector)
| |
| '-- [ Amazon DynamoDB ] (single-table vector scan & match)
|
'-- 2. Read Document / View Slide
|
v
[ S3 Presigned URL ] --> Instant byte-range streaming

Why Serverless?

Universities experience extreme traffic patterns: virtually 0 users during summer break, and thousands of concurrent users during exam week.
By building NotesForAll AI with AWS Lambda, Amazon DynamoDB, and Amazon S3, the infrastructure scales effortlessly from zero to peak demand with:
  • $0 idle operational cost.
  • No servers to patch or maintain.
  • A single-table NoSQL schema that retrieves files, summaries, and FAQs in single-digit milliseconds.



Key Takeaway

Building NotesForAll AI proved that you don't need a bloated monolithic backend or a complex dedicated vector cluster to build real-world, high-performance AI applications.
By combining Amazon Bedrock's foundation models with the elasticity of AWS Serverless, you can turn static, forgotten courseware into an interactive study companion that genuinely helps students learn faster.
Next time exam season rolls around, leave  Ctrl+F  in the past. 🚀
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article