About Pora Mama (পড়া মামা)

Pora Mama is an AI-powered personal study companion built for students who are drowning in scattered study material — handwritten notes, class slides, textbook chapters — and need one warm, reliable friend to turn all of it into an actual exam plan.

The name means "study uncle/friend" in Bangla — deliberately informal, the opposite of a cold productivity tool. The goal is for it to feel like a supportive senior who organizes your notes, tells you what to study and when, quizzes you, and honestly tells you where you still stand.

The problem it solves

Most students preparing for an exam have their material in three incompatible places — a photo of handwritten class notes, a PDF of slides, and a textbook chapter — with no time to merge them, no clear read on how "ready" they actually are, and no easy way to test themselves the way a real exam will.

How it works — six stages

  1. Note Collector — upload handwritten notes, slides, and book pages; OCR extracts the text and the student corrects any mistakes before it becomes "the" note.
  2. Note Maker — merges every source into one clean, organized study note per topic, with definitions, formulas, worked examples, and a "common exam traps" section.
  3. Video Finder — suggests targeted YouTube search queries per subtopic, so the student finds real explainer videos instead of the app inventing fake links.
  4. Planner — builds a day-by-day study schedule from the days remaining and daily hours available, with lighter revision days near the end.
  5. Examiner — generates a mock written exam matched to university-level difficulty.
  6. Khata Checker — the student photographs their handwritten answer script ("khata"); the app grades it against a rubric, gives per-question feedback, and produces a final readiness report with weakest topics ranked.

Design principles

  • A friend, not a black box. Every AI output is editable and regenerable — nothing is presented as final without a way to correct it.
  • Bilingual by default. Bangla-first UI with English technical terms kept in English, plus a language toggle.
  • Never fabricate. No invented video titles, links, or scores — low-confidence outputs are marked as uncertain rather than guessed.
  • Mobile-first. Most real use happens on a phone right after handwriting a mock answer.

Technical approach

The current build target is Google AI Studio, using Gemini's multimodal input for OCR across handwritten photos, slides, and PDFs. Each of the six stages is built as a separate, structured-output function rather than one giant prompt — so the app is already shaped like a set of agents, even before any agent framework is involved.

The longer-term architecture layers in:

  • RAG — chunked, embedded study sources for topic-level retrieval instead of stuffing everything into context
  • LangGraph — an orchestration graph where each stage is a node, sharing one session state object
  • MCP — external capabilities (YouTube search, calendar reminders, file storage) exposed as swappable tool servers

The plan is to validate the full six-stage experience in AI Studio first, then migrate only the stages that actually need retrieval-quality upgrades — most likely Note Maker and Examiner.

Hardest problem

Grading a photographed handwritten answer script accurately — reading messy handwriting and judging partial credit fairly — is harder than every other stage combined, and it's the stage most worth stress-testing early, even though it comes last in the flow.

Status

  • Structured AI Studio build prompt: written
  • Tagline: "পড়া মামা — study smart, not alone."
  • Thumbnail illustration: done
  • Persona/system-prompt voice, session-state schema: not yet started

Built With

Share this project:

Updates

Submission history