Adaptive university exam-learning platform for TKM College of Engineering (KTU 2024 Scheme, Semesters S3–S8). Transforms official syllabus, previous-year questions (PYQs), student mastery, mistakes, and study time into personalized, high-yield learning sessions.
-blue?style=flat-square&logo=typescript){kind=icon}
Engineering students typically have all the raw materials and none of the direction:
- A massive official university syllabus with no guidance on high-yield revision
- Scattered lecture notes that don't reflect actual KTU exam mark distributions
- Previous-year questions (PYQs) disconnected from specific syllabus topics
- Illusions of competence: topics students think they know but consistently fail in exams
- Critical time scarcity during semester internals and university finals
Nobody tells students what to study next. They guess, passively re-read familiar material, or avoid their weakest modules entirely.
Modulr replaces that guesswork with a deterministic decision engine. It evaluates the student's syllabus coverage, mapped PYQs, current mastery level, unresolved mistakes, and spaced revision schedule to recommend exactly one single action with a transparent pedagogical reason.
The engine implements an active, evidence-driven learning loop:
┌─────────────────────────────────┐
│ ACADEMIC SYLLABUS │
│ (KTU 2024 Scheme S3–S8) │
└────────────────┬────────────────┘
│
▼
┌──────────────┐ ┌──────────────┐ ┌─────────────────────┐
│ EXAM PYQS │ ───────► │ DECISION │ ◄─────── │ LEARNING STATE │
│ (Topic Mapped) │ ENGINE │ │ (0–6 Mastery / Fix) │
└──────────────┘ └───────┬──────┘ └─────────────────────┘
│
One Defensible Next Action
│
┌────────────┬───────────┼───────────┬────────────┐
▼ ▼ ▼ ▼ ▼
[ TEACH ] [ RECALL ] [ PRACTICE ] [ FIX ] [ EXAM ]
Zero-to-Pro Spaced PYQ Drill Mistake Day-Before
Foundation Revision Repair Attack Mode
│ │ │ │ │
└────────────┴───────────┼───────────┴────────────┘
│
▼
┌─────────────────────┐
│ TIMED STUDY PLAN │
│ (15 / 30 / 45 / 60) │
└──────────┬──────────┘
│
▼
┌─────────────────────┐
│ CONTEXTUAL PROMPT │
│ + 1-Click Launch │
│ (ChatGPT/Claude/Gem)│
└──────────┬──────────┘
│
▼
┌─────────────────────┐
│ EVALUATED EVIDENCE │
│ (Strong / Partial / │
│ Struggled) │
└──────────┬──────────┘
│
▼
┌─────────────────────┐
│ MASTERY UPDATE │
│ (0–6 Integer) │
└─────────────────────┘
The loop is closed: every study session ends in concrete recorded evidence. Evidence drives mastery, and mastery updates the next recommendation.
- Pure Function: Given the same academic state, it yields the exact same recommendation. No hidden clock dependencies, no stochastic drift, and no random suggestions.
- Five Specialized Task Types:
TEACH– Structural concept introduction using the Zero-to-Pro teaching chain for unstudied topics.RECALL– Active spaced retrieval for concepts whose memory half-life has elapsed.PRACTICE– Targeted PYQ applications for concepts started but not yet solidified.FIX– High-priority diagnostic sessions specifically targeting recorded student mistakes.EXAM– High-yield university exam attack plan when a subject is broadly assessed.
- Transparent Reasoning: Every recommendation explicitly tells the student why it was chosen (e.g., "You made 2 mistakes in Module 2 that require fixing before moving forward").
Integrated directly into every subject page for high-stakes, time-constrained revision:
- 5 Adaptive Time Brackets:
- < 1 Hour: Ultra-high-yield recall only (critical definitions, core formulas, must-know traps).
- 1–3 Hours: Essential concepts + top recurring KTU question patterns + scoring edge cases.
- 3–6 Hours: High-yield syllabus roadmap + prioritized PYQ drill + targeted weak module repair.
- 6–10 Hours: Comprehensive high-yield attack plan + exam-format practice + concept reconstruction.
- 10+ Hours: Deep high-yield mastery + full simulated mock test + final memory consolidation.
- Priority Scoring Engine: Classifies modules into Tier A (Must-Pass High Yield), Tier B (Core Standard), and Tier C (Buffer/Niche) based on formula density, exam focus weightage, definition density, and mapped PYQ volume.
- Built-in Session Timer: Active-phase countdown timer with visual pulsing status, pause/resume controls, and completion sound alerts.
- 1-Click External AI Launchers:
- ChatGPT: Direct launch via
https://chatgpt.com/?q=...with pre-filled prompt. - Claude: Direct launch via
https://claude.ai/new?q=...with pre-filled prompt. - Gemini: Automated clipboard copy with instant link to
https://gemini.google.com/app.
- ChatGPT: Direct launch via
- Popup-Safe Navigation: Uses native anchor navigation to prevent browser popup blockers from interrupting the study flow.
A curated library of 30 core computer science disciplines and 200+ topics, engineered around a 9-stage pedagogical pipeline:
MAP → PRIORITIZE → LEARN → APPLY → RECALL → CHECK → REPAIR → CONNECT → MASTER
- Experience Tiers: Beginner (mental models, zero jargon), Intermediate (high-yield per minute, core implementations), and Advanced (systems intuition, architectural tradeoffs, edge cases).
- Targeted Goals: Concept Understanding, University Exams, Technical Interviews, Problem Solving (DSA), and Real-World Projects.
- Pedagogical Styles: Simple Intuition, Socratic Dialogue, Question → Discovery, Visual/Spatial ASCII Models, Code-First Tracing, and Rigorous Theory-First.
Modulr intentionally runs no proprietary server-side LLMs. Instead, it generates context-rich, provider-agnostic study prompts for students to use with their favorite frontier models:
- Zero API Costs & Zero Keys: Students don't need API subscriptions, tokens, or configuration.
- 100% Inspectable Prompts: Students see and control the exact academic instructions given to the AI.
- Official Syllabus Verification Instruction: Prompts mandate that the external AI verify facts against official TKM College of Engineering records (
https://tkmce.ac.in/) when uncertain. - PYQ Honesty Directive: Prevents AI hallucinations by strictly distinguishing verified KTU previous-year questions from synthesized practice questions.
Mastery is an integer scale from 0 to 6 that is deliberately conservative and difficult to move:
- Opening a lesson awards 0 mastery. It records exposure only; mastery remains
null. - Reading a note does not equal mastery.
TEACHsessions register exposure, not verified proof. - Evidence drives mastery:
strongoutcome (correct without hints) → increments mastery.partialoutcome (needed prompting or partial recall) → minor adjustment.struggledoutcome (fundamental confusion) → decreases score and schedules aFIXsession.
- Different evidence weights: Active recall and verified PYQ success move mastery significantly faster than simple self-assessment checks.
- Honest Unassessed State: Displayed as
0% (Unassessed)— not failed.
- Course codes are not globally unique in the KTU 2024 scheme (e.g.,
24CSP304is Algorithms in CSE, but Data Structures & Algorithms in CSE [AI]). - Academic identity is strictly program-scoped:
(programId, subjectCode)with stable namespaced identifiers across all routes, storage, and cross-links.
Full curriculum representation across 259 subjects spanning Semesters S3 through S8:
| Program | Stream Name | Subjects | Route Slug |
|---|---|---|---|
| CS | Computer Science & Engineering | 108 | /syllabus/cse |
| CS_AI | Computer Science & Engineering (Artificial Intelligence) | 113 | /syllabus/cse-ai |
| ER | Electrical & Computer Engineering | 38 | /syllabus/er |
| Total | KTU 2024 Scheme (S3–S8) | 259 |
flowchart TD
subgraph DataLayer [Data Layer - Build-time Static]
A[Official KTU Syllabus] --> B[Syllabus Parser]
B --> C[Curriculum Registry]
D[PYQ Repository] --> E[Token Overlap Matcher]
E --> F[Mapped PYQs]
end
subgraph StateLayer [Client-Side State - localStorage]
G[Topic Mastery 0-6]
H[Open Mistakes Ledger]
I[Revision-Due Timestamps]
J[Session History]
end
subgraph Engine [Core Engines]
C --> K[Decision Engine]
F --> K
G --> K
H --> K
I --> K
K --> L{Recommended Action}
L -->|Unstarted| M[TEACH]
L -->|Due| N[RECALL]
L -->|Weak| O[PRACTICE]
L -->|Mistake| P[FIX]
L -->|Assessed| Q[EXAM]
L --> R[Session Planner]
R --> S[Contextual Prompt Engine]
end
subgraph Execution [Active Study Execution]
S --> T[1-Click AI Launchers]
T --> U[ChatGPT / Claude / Gemini]
R --> V[Live Study Timer]
U --> W[Study & Self-Evaluation]
V --> W
W --> X[Session Evidence Logger]
end
X --> G
X --> H
X --> I
X --> J
UI Layer app/ (App Router), components/subject/, components/navigation/
│
Learning Engine lib/learning/ (decision.ts, session.ts, mastery.ts, state.ts)
│
Prompt Engine lib/learning/prompts/, lib/prompts/dayBeforeExam.ts, lib/learn-cs/ai.ts
│
Academic Data lib/syllabusData.ts, lib/content.ts, lib/pyqs.ts, lib/notes/
│
Domain Foundations lib/domain.ts, lib/types.ts, lib/urls.ts
| Component | Technology | Version | Purpose |
|---|---|---|---|
| Framework | Next.js (App Router) | 14.2.35 |
Static Site Generation (1607 pre-rendered HTML pages) |
| UI Library | React | ^18.3.1 |
Component composition and reactive state |
| Language | TypeScript | ^5.5.4 |
Strict typing, shared domain models, zero any |
| Styling | Tailwind CSS | ^3.4.7 |
Utility styling with semantic CSS variable design tokens |
| Testing | Vitest | ^2.1.9 |
180 unit, integration, and idempotency tests across 16 suites |
| Linting | ESLint | ^8.57.1 |
Code standards via eslint-config-next |
| Persistence | Browser localStorage |
Native | 100% private client-side storage (no tracking, no DB) |
| Deployment | Vercel / GitHub Pages | CI/CD | Static global edge deployment |
Every commit and pull request must satisfy rigorous static verification:
npm test # Vitest: runs all 180 tests across 16 test files
npm run lint # ESLint: checks Next.js and React best practices
npx tsc --noEmit # TypeScript: verifies strict type safety across all files
npm run build # Production build: verifies SSG for all 1607 static pagesnpm run validate:content # Validates subject registries, branch isolation, and note counts
npm run validate:syllabus# Validates syllabus structures, prerequisite DAGs, and cross-links
npm run import:syllabus # Regenerates lib/syllabusData.ts (strictly idempotent)Note on Importer Idempotency: Running
npm run import:syllabustwice consecutively produces byte-identical output. A dedicated Vitest test (tests/import-idempotent.test.ts) asserts this to prevent silent data regressions.
TKM-notes/
├── app/
│ ├── page.tsx # Landing & Quick Navigation
│ ├── syllabus/ # Browse by Program (CSE, CSE-AI, ER) & Semester
│ │ └── [program]/[semester]/[subject]/ # Subject view & Day-Before Exam Mode
│ ├── ai-study/ # AI-guided study session hub
│ ├── learn-cs/ # 30-subject Computer Science curriculum & roadmap
│ ├── revision/ # Night-before exam revision assistant
│ ├── planner/ # Semester exam planner & timetable tracker
│ ├── pyqs/ # Previous year question browser & topic filter
│ └── prompt-lab/ # Advanced prompt configuration workbench
├── components/
│ ├── subject/ # LearnSession, DayBeforeExamCard, ExternalAiHandoff
│ ├── prompt-lab/ # Prompt builder and template customizer
│ └── navigation/ # AppHeader, SubjectTabs, Breadcrumbs
├── lib/
│ ├── learning/ # Decision engine, session planner, mastery calculator
│ ├── prompts/ # Day-before exam prompts, zero-to-pro templates
│ ├── learn-cs/ # Learn CS AI generator, topic DAG, categories
│ ├── domain.ts # Canonical academic hierarchy & program definitions
│ └── syllabusData.ts # Static typed syllabus data for 259 subjects
├── scripts/ # Syllabus import and data integrity verification scripts
└── tests/ # 16 Vitest test suites (180 automated tests)
- Canonical domain model with program-scoped academic identity (
domain.ts) - Full KTU 2024 scheme syllabus coverage (259 subjects across CS, CS_AI, ER)
- Deterministic decision engine (
decideNextAction) across all five learning tasks - Time-boxed session planner (
planSession) with minute-exact allocations - Evidence-weighted 0–6 mastery state engine with mistake tracking (
FIX) - Day-Before Exam Attack Mode with 5 time brackets and module priority scoring
- Interactive live study countdown timer with sound notifications
- 1-Click AI launchers (ChatGPT, Claude, Gemini) with popup-safe navigation
- Learn CS catalog (30 subjects, 200+ topics) with high-yield pedagogical prompts
- Rigorous verification suite: 180 passing Vitest tests, clean lint, clean TypeScript check
- No Embedded LLM Server: We deliberately do not run a hosted LLM backend or proxy API calls. This guarantees zero API costs, zero telemetry, total privacy, and prevents API token exhaustion for students.
- No User Accounts / Cloud Database: All state resides securely in the student's browser
localStorage. - No Artificial Gamification: No streaks, badges, or animated confetti. The interface respects the student's cognitive focus.
- State Import / Export: Allow students to export their
localStoragelearning state to a JSON file and restore it on another device. - Expanded Written Notes: Expand handwritten / curated notes coverage beyond the initial 28 core subjects (the learning engine already functions for all 259 subjects via syllabus and PYQs).
- Graded PYQ Confidence: Upgrade token-overlap PYQ mapping to output graded confidence tags (
high,medium,suggested). - Offline PWA Support: Add service workers and offline caching for note reading during network dropouts.
Sreerang — TKM College of Engineering
Repository: cosmiccoder200x-sys/TKM-Notes