Turn Any Video Into A Personal Learning System.
Saras AI transforms educational content into structured lessons, grounded recall, personalized knowledge maps, adaptive revision, exam plans, and practical projects.
The Saras Intelligence Loop
From Content to Mastery
Watching a video is passive. Saras wraps video content into an active, verified loop: Watch → Understand → Verify → Remember → Connect → Revise → Prepare → Build.
01 / Turn a Video Into Structure
Paste any YouTube URL and Saras transforms the raw caption stream into meaningful topic chapters with timestamp boundaries, caching them permanently for deterministic recall.
02 / Don't Just Watch. Verify.
Saras automatically pauses playback at chapter boundaries and challenges you with diagnostic checkpoint questions. Future video chapters remain locked until understanding is verified.
What role does an activation function serve in an artificial neuron?
03 / Turn Watching Into Memory
Every video generates persistent Smart Notes organized into Overview, Key Concepts, Definitions, and Common Mistakes alongside interactive spaced repetition flashcard decks.
Key Concepts & Takeaways
Synthesized key definitions linked directly to source video timestamps.
Active Spaced Recall
Drill definitions, toggle mastery, and review cards directly in Flashcards Studio.
04 / See What You Actually Know
Saras connects concepts across all your completed lessons into an interactive, hierarchical Knowledge Map. Colors clearly reflect your verified retention: Lime for Mastered, Lavender for Developing, Coral for Needs Review, and Cream for Not Studied.
05 / Saras Chooses What You Should Revise
Start an adaptive revision session and Saras automatically curates questions targeting weak concepts, recent mistakes, and older material using a deterministic ranking formula.
06 / Turn Knowledge Into a Study Plan
Tell Saras what you're preparing for, when your exam is, and how much time you have. Saras topologically schedules prerequisites first, guarantees daily time budgets, and dynamically replans if mistakes occur.
Machine Learning Midterm
7 Days Remaining · 60m / Day07 / Learn Something. Then Build It.
When a technical video concludes, Saras transforms the concepts into an actionable software project with requirements and milestone tracking grounded directly in taught skills.
Neural Network Simulator in Python
The Saras Learning Cycle
Every lesson strengthens recall, updates your map, and powers personalized revision.
Watch
Stream video chapters with timestamped captions
Understand
Read AI-synthesized Smart Notes with key definitions
Verify
Solve checkpoint quizzes before unlocking next part
Recall
Drill active recall cards with spaced repetition
Connect
Explore knowledge graph & prerequisite relationships
Revise
Run personalized revision sessions on weak concepts
Apply
Build grounded projects mapped directly to video skills
Master
Track readiness toward your exam target date
Why Specialized Agents?
One unified pipeline with 8 distinct, deterministic jobs. Rather than relying on a single monolithic prompt, each agent solves one specific pedagogical task.
Ingestion Agent
Extracts video ID, checks deterministic SQLite/Prisma cache, and parses timestamped transcript tokens for downstream LLM processing.
Segmentation Agent
Applies pedagogical prompt heuristics to detect semantic topic transitions and partition raw video captions into structured chapters.
Quiz Agent
Synthesizes 4-option diagnostic multiple-choice questions strictly from the current video segment, gating progress until answered correctly.
Recall Agent
Extracts core definitions and active recall pairs directly taught in the video and structures them into persistent study decks.
Concept Engine
Identifies atomic technical concepts and extracts prerequisite and hierarchical relationships to construct a data-driven Knowledge Map.
Revision Engine
Ranks revision candidates using a multi-factor priority formula (weakness, mistake frequency, age, recency) to curate focused study runs.
Exam Planner
Applies topological sorting on concept prerequisites to build a strict daily study timetable that automatically replans upon performance changes.
Project Generator
Transforms technical concepts actually taught in the video into a grounded software project roadmap with requirements and milestone tracking.
Deep Focus Engine
Analyzes face landmarks entirely on-device with MediaPipe & WebAssembly. Automatically pauses playback if gaze leaves the viewport.
Live Pipeline Execution
Simulate the 8-stage transformation from YouTube video URL to grounded learning curriculum.
* Note: Sample metrics above are an illustrative simulation of pipeline outputs.
The Core Tech Stack
Built on production-grade web technologies, local in-browser vision models, and dual-provider Gemma 4 intelligence.
Next.js 14
Hybrid SSR & React Server Actions
Tailwind CSS
Neo-brutalist custom tokens
Prisma + SQLite
Normalized relational storage
Gemma 4 Client
Local Ollama & Cloud Gemini
YouTube IFrame
Precision timestamp gating
MediaPipe WASM
100% on-device attention
100% On-Device Attention Tracking
Deep Focus runs entirely inside your browser canvas using WebAssembly. Camera frames are processed in-memory and discarded instantly. Video is never streamed, recorded to disk, or sent to any server.
Privacy Trust Guarantee
Product Implementation Status
Video Ingestion & Deduplication
Transcript extraction and SQLite caching
Segmentation & Checkpoints
Topic-bounded chapters with quiz gating
Smart Notes & Flashcards
Structured notes and active recall decks
Hierarchical Knowledge Map
Domain clustering and prerequisite DAG
Adaptive AI Revision Engine
Smart mix and weak area candidate selection
Adaptive Exam Mode
Prerequisite study roadmap with replanning
Deep Focus Studio
Local attention monitoring & auto-pause
Video → Project Generator
Grounded software project specs
Multi-User Isolation
Scoped progress, mastery, and exam plans
Don't Just Watch.
Build Knowledge.
Give Saras a video. Let it structure the lesson, challenge your understanding, map what you know, help you revise, prepare, and build.