AWS Builder Center
bodh. - Making learning accessible for everyone

bodh. - Making learning accessible for everyone

Build a learning system where language is not a barrier, confusion is not a dead end, and every student gets a path that adapts to them.

Advancing with Full Stack Applications and System Design

Building bodh.: An AI-Powered Bilingual Learning Platform for Rural India

How we used Amazon Bedrock, Amazon S3, Amazon DynamoDB, and AWS AgentCore to build a personalized Computer Science learning experience

Learning Computer Science is hard enough in a language you're fluent in. Learning it in a language you're still catching up on makes it harder still.
For students in rural and underserved parts of India, that gap isn't just about vocabulary. It's about whether a lesson makes sense the first time, whether help arrives before frustration sets in, and whether the platform still works when the connection doesn't.
That's the problem we set out to explore while building bodh. during the Bharat Builds Tour.
What if a learning platform adapted not just to what a student wants to learn, but to how they learn best — and kept working even when the internet didn't cooperate?
That question became the foundation of bodh.

The idea behind bodh.

bodh. is a bilingual, AI-powered learning platform, starting with Data Structures and Algorithms for students aged 15–20. The first version supports English and Hindi, covering:
  • Arrays
  • Linked Lists
  • Stacks
  • Queues
  • Binary Search
  • Recursion
Bilingual content alone wasn't the interesting part. We didn't want to just translate an English article into Hindi and call it done. Translation solves a vocabulary problem, not a comprehension one — and if a student doesn't understand an explanation, handing them the identical explanation in a different language doesn't actually fix anything.
So we built bodh. around a single loop:
Learn → Practice → Assess → Diagnose → Recommend → Learn Again
Every part of the system exists to keep that loop running around the student, not around the content.
Rural connectivity was the other constraint we designed around from day one. bodh. ships with offline-first seed content — lessons load straight from local content with zero AWS configuration, so the app never breaks for a student on a patchy connection. AI, personalization, and progress tracking layer on top when the network's there; they were never a precondition for a student to start learning.

From a Content Platform to a Learning System

Most learning platforms follow a simple path: content, then quiz, then score. bodh. runs the loop above instead, with each stage doing real work rather than just passing through to the next.
Take a student learning Binary Search. They read the lesson in Hindi, then take a five-question quiz — and score 40%.
A typical platform stops there: pass or fail. bodh. treats that score as a starting point. It analyzes which questions the student missed and surfaces the specific sub-concept behind the pattern — in this case, the student understood the idea of halving the search space but kept miscalculating the new boundaries after each comparison.
At that point, the student can ask the AI directly: "Explain this differently."
Instead of replaying the same walkthrough, bodh. generates a new one — trading step-by-step index arithmetic for a physical analogy, like flipping through a phone book by always jumping to the middle page and discarding the half that can't contain the name. The student practices again, and the new attempt becomes part of their learning history — data the system uses the next time it recommends what to study.
That's the shift: the feedback loop runs around the student, not the content.

Building the Architecture on AWS

One of the most important decisions we made early was to keep the architecture modular. The application layer owns the user experience and the learning workflow. AWS services provide the infrastructure underneath — AI, content, and learner data — each with one clear job, instead of one large, undifferentiated AI workflow doing everything.


Amazon Bedrock as the AI Layer

Amazon Bedrock is the core intelligence layer of bodh., handling:
  1. Teaching. A student requests an explanation of a concept in their chosen language. The application passes Bedrock the topic, language, and learning context; Bedrock generates the explanation while the application controls the surrounding learning experience.
  1. Adaptive explanation. The "explain this differently" interaction above. Instead of replaying the same content, Bedrock generates a genuinely different explanation — a new analogy, a different entry point, a different level of abstraction. This is where the AI stops being a content generator and becomes part of the teaching loop itself.
  1. Quiz generation. Each quiz is generated around a single topic — five questions for the MVP — tightly scoped to the concept the student just learned.
  1. Learning analysis. Bedrock combines a student's quiz performance with their learning history to identify which concepts need more attention.
  1. Recommendations. The system uses that analysis to recommend what a student should learn or practice next.
Where AI stops and logic starts. Not every part of this needed to be AI-driven — and deciding where to draw that line was one of the more consequential calls we made. Scoring and mastery thresholds run on deterministic application logic: the same quiz result produces the same score, every time, with no model variance to account for. Bedrock is reserved for what actually needs natural-language reasoning — generating an explanation, reading a pattern of wrong answers, writing feedback that sounds like it's talking to this student. The AI doesn't need to control everything to be central to the product; it needs to control the right things.

Amazon S3 as the Content Layer

Educational content changes far less often than student activity does, so we kept it out of the application code entirely and stored it in Amazon S3, organized by topic and language:
articles/<topic>/<lang>.json → one lesson per topic per language
mindmaps/<topic>.json → one visual concept map per topic

That separation means adding a new topic or a new language is a content change, not a code change — and it's what makes the offline fallback possible in the first place: the same structure that lives in S3 for production also ships as seed JSON on disk, so the app has something to read from even with zero AWS connectivity.


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