CHAIN 239020.XYZ

JiYan AI Tutor

Full-Subject Postgraduate Exam AI Tutor & Contextual Hybrid RAG Platform

A full-stack AI tutoring platform covering 24+ postgraduate exam disciplines. Indexes 61 authoritative textbooks into a parent-child hybrid RAG corpus inside a single-screen 3-tab notebook with 9 studio tools and zero-spoiler Socratic guidance.

  • Vue 3
  • Vite 6
  • FastAPI
  • SQLite (WAL)
  • ChromaDB
  • DeepSeek
  • Qwen Vision
  • ECharts
  • KaTeX
Plate
PLATE 01
Role
Primary Author
Status
Live in Production · Active
Year
2026
Scope
Primary Author: Independently architected the full-stack application, Contextual Hybrid RAG pipeline, Socratic tutoring chain, and production deployment.
PLATE 01 · DECK 4 SLIDES

Interactive HTML Slide Deck

Navigate with controls, keyboard arrows, slide indicators, or Fullscreen presentation mode.

PLATE 01 ACADEMIC OVERVIEW
SLIDE 01 / 04

24+ Exam Disciplines & Syllabus Command Console

jiyan.ink · ACADEMIC CONSOLE24+ SUBJECTS
408 CS306 Med311 Edu312 Psych199 MgmtMath / Politics / English
408 Data Structures88%
Operating Systems76%
Calculus & Probability91%
Computer Networks82%
24+Exam Disciplines
61Indexed Textbooks
36,464Parent Blocks
154,490Child Chunks
MASTERY VELOCITY · ECHARTSKaTeX · WAL
Covers 24+ unified and public exam subjectsECharts mastery tracking and countdown pacingFast Vue 3 + Vite 6 + KaTeX math rendering
SLIDE 01 / 04
Open HTML Deck

Consolidates 24+ exam syllabi, revision cadence, and concept mastery metrics into a focused academic dashboard.

  • Covers 24+ unified and public exam subjects
  • ECharts mastery tracking and countdown pacing
  • Fast Vue 3 + Vite 6 + KaTeX math rendering
  • 24+ Exam Disciplines
  • 61 Indexed Textbooks
  • 154,490 Child RAG Chunks
  • 9 Studio Learning Tools

Context & Engineering Constraints

General-purpose LLMs answering postgraduate exam questions frequently lack textbook grounding, hallucinate syllabus details, and spoil full solutions before students build independent reasoning habits.

System Architecture

  1. 01

    Parent-Child Contextual Hybrid RAG

    Segments 61 authoritative textbooks with contextual headers into 36,464 parent blocks and 154,490 fine-grained child chunks retrieved through ChromaDB vector and lexical fusion.

  2. 02

    FastAPI + SQLite (WAL) Multimodal Backend

    FastAPI orchestrates DeepSeek reasoning alongside Qwen Vision OCR for handwritten/printed questions, backed by SQLite in WAL mode for concurrent mastery tracking.

  3. 03

    Single-Screen 3-Tab Notebook & 9-Tool Studio

    Vue 3 and Vite 6 power a unified Materials / Dialogue / Studio workspace with native KaTeX math rendering, ECharts mastery analytics, and 9 study generators.

Engineering Highlights

  • 61 authoritative textbooks indexed via parent-child chunking (36,464 parent / 154,490 child chunks) so every concept citation traces back to exact syllabus chapters.
  • Engineered a Zero-Spoiler Socratic chain that isolates problem cues, asks guided prompts first, and reveals official grading rubrics only after student derivation.
  • Nine integrated studio tools convert study materials into flashcards, mind maps, LaTeX proofs, micro-lectures, and practice quizzes inside one screen.
  • Optimized for intensive exam-prep sessions using streaming responses and SQLite WAL persistence for predictable latency.