Spaces:

iamfebin
/

algospaced-dsa

Sleeping

App Files Files Community

algospaced-dsa / System_Design_DSA_Learn.md

iamfebin

Configure AlgoSpaced for Hugging Face Spaces deployment

aaa634c 27 days ago

preview code

Raw

History Blame Contribute Delete

16.1 kB

# System Design Document: AlgoSpaced

A cloud-native, mobile-responsive Spaced Repetition System (SRS) designed to help software engineering candidates master the LeetCode 150 patterns. Rather than rote memorization, AlgoSpaced optimizes the retention of algorithmic paradigms using dynamic intervals, cloud syncing, an automated LLM Code Reviewer, a gamified Daily Streak System, and an automated Google Drive Ingestion Pipeline.

1. System Architecture Overview

The system is designed with a lightweight, cross-platform approach to ensure seamless transition between desktop (IDE practice) and mobile (on-the-go review) views.

+--------------------------------------------------------------+
|                        Client Tier                           |
|      (Mobile Browser / Desktop Browser - Streamlit App)      |
+------------------------------+-------------------------------+
                               |
                        HTTPS (JSON / REST)
                               |
                               v
+------------------------------+-------------------------------+
|                        Application Tier                      |
|                  (Python Serverless/Streamlit)                |
+-------+----------------------+-----------------------+-------+
        |                      |                       |
        | Secure API           | Sync Data             | Database Queries
        | (gemini-2.5-flash)   | Google Drive & Docs   | (SSL/TLS Connection)
        v                      v                       v
+-------+--------------------+ +--------------------+ +-------+-------+
|        External AI Engine  | | Google Workspace   | | Database Tier |
|  (Google Gemini API)       | | (Drive & Docs APIs)| | (Supabase/PG) |
+----------------------------+ +--------------------+ +---------------+

Components

Client Tier (Frontend): A unified, responsive web interface built using Streamlit. It automatically handles fluid widths and touch-friendly target sizes for mobile screens, adapting beautifully between viewports.
Application Tier (Backend Logic): Python logic running the core Spaced Repetition engine (calculating intervals), processing user code inputs, calculating daily active streaks, coordinating Google Drive folder scans, and structuring payloads for the LLM evaluation.
External AI Engine (Gemini API): An intelligent grading agent utilizing gemini-2.5-flash-preview-09-2025 to evaluate candidate code for logical correctness, time/space complexity, and pattern edge-cases.
Database Tier (Cloud Database): A managed PostgreSQL instance (e.g., Supabase) storing user progress, problem history, review queues, and user streak statistics.

2. Spaced Repetition & Streak Algorithms

2.1 Leitner System Spaced Repetition

To avoid linear burnout, AlgoSpaced implements a customized Leitner System optimized for algorithmic patterns. Problems transition between 5 "Boxes", where each box represents a distinct revision interval.

Let the next review date $D_{next}$ for a problem $p$ be defined as:

$D_{next} = D_{last} + I(b)$

Where $D_{last}$ is the date of the last review, and $I(b)$ is the interval (in days) associated with the problem's current Box level $b \in {1, 2, 3, 4, 5}$:

$I(b) = \begin{cases} 1 \text{ day} & \text{if } b = 1 \text{ (Forgot / Hard)} \\ 3 \text{ days} & \text{if } b = 2 \\ 7 \text{ days} & \text{if } b = 3 \\ 14 \text{ days} & \text{if } b = 4 \\ 30 \text{ days} & \text{if } b = 5 \text{ (Mastered)} \end{cases}$

2.2 SRS State Transitions

On Correct Evaluation (Easy / AI Approved):

$$b_{new} = \min(b_{current} + 1, 5)$$
On Mixed Evaluation (Medium / Minor Bugs):

$$b_{new} = b_{current}$$

(No change in Box level, but resets the countdown interval to $I(b_{current})$ days from today)
On Incorrect/Forgotten Evaluation (Hard):

$$b_{new} = 1$$

(Resets to Box 1 immediately to build retention)

2.3 Daily Streak Mathematical Model

To keep motivation high, the system maintains a running daily streak tracking active participation. A streak increments when the user completes at least one practice session (submitting code or passing a review) within a calendar day.

Let $T_{now}$ be the current timestamp in the user's localized timezone, and let $Date(T)$ be the calendar date portion of a timestamp. Let $D_{last_active}$ be the date of the user's last logged review session.

When a review session is completed, the system evaluates the state of the streak:

$\text{Streak Status} = \begin{cases} \text{No Change} & \text{if } Date(T_{now}) = D_{last\_active} \\ \text{Increment Streak} & \text{if } Date(T_{now}) = D_{last\_active} + 1 \text{ day} \\ \text{Reset Streak to 1} & \text{if } Date(T_{now}) > D_{last\_active} + 1 \text{ day} \end{cases}$

Lazy Streak Verification on Login

To ensure the visual streak updates immediately even before the user starts a session, the frontend runs a checking routine whenever the application loads:

$\text{Current Streak} = \begin{cases} 0 & \text{if } Date(T_{now}) > D_{last\_active} + 1 \text{ day} \\ \text{Unchanged} & \text{otherwise} \end{cases}$

3. Google Drive Ingestion Pipeline

To make registration friction-free, users only need to drop their LeetCode notes as Google Docs into a dedicated Google Drive folder (e.g., named LeetCode-150-Reviews). The system will parse and sync them automatically.

3.1 Authentication Workflow

The background sync worker uses a secure Google Cloud service account with specific read-only permissions:

Scope Required: https://www.googleapis.com/auth/drive.readonly and https://www.googleapis.com/auth/documents.readonly

3.2 Document Structured Parsing Rules

Your notes are structured inside the Google Doc in a human-friendly format. The ingestion script parses the Google Doc's JSON structural elements (body.content) and applies regex matching to dynamically extract fields:

+--------------------------------------------------------------+
|                     Google Doc Layout                        |
+--------------------------------------------------------------+
| Document Title: [Problem Name] (e.g. "Two Sum")              |
|                                                              |
| Pattern: [Pattern Name]                                      |
| Complexity: Time: [O(N)], Space: [O(N)]                      |
|                                                              |
| ---                                                          |
| [Optional: Aha! Moment Line / Key insight]                  |
|                                                              |
| Code:                                                        |
| ```python                                                    |
| ... python code solution ...                                 |
| ```                                                          |
+--------------------------------------------------------------+

Problem Name: Extracted directly from the Doc filename.
Pattern Name: Found via Regex search for line: Pattern:\s*(.*). Default to Unassigned if not found.
Complexities: Found via Regex matching: Time:\s*(\S+)\s*,\s*Space:\s*(\S+).
Correct Working Code Block: Parsed by grabbing all texts following a Code: delimiter or inside formatted monospace blocks.

4. Data Models & Database Schema

The database must be hosted in the cloud to sync perfectly between your phone and laptop. Below is the relational schema designed for PostgreSQL.

PostgreSQL Schema Initialization Script

-- Enable UUID extension
CREATE EXTENSION IF NOT EXISTS "uuid-ossp";

-- Table: problems
CREATE TABLE problems (
    id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
    name VARCHAR(255) NOT NULL UNIQUE,
    pattern VARCHAR(100) NOT NULL,
    difficulty VARCHAR(20) NOT NULL DEFAULT 'Medium' CHECK (difficulty IN ('Easy', 'Medium', 'Hard')),
    google_doc_url TEXT,
    google_doc_id VARCHAR(100) UNIQUE, -- Tracks file to prevent duplicates
    optimal_time VARCHAR(50) NOT NULL DEFAULT 'O(N)',
    optimal_space VARCHAR(50) NOT NULL DEFAULT 'O(N)',
    reference_code TEXT, -- Stores reference code synced from Doc
    created_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP
);

-- Table: user_reviews
CREATE TABLE user_reviews (
    id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
    problem_id UUID NOT NULL REFERENCES problems(id) ON DELETE CASCADE,
    box_level INTEGER NOT NULL DEFAULT 1 CHECK (box_level BETWEEN 1 AND 5),
    last_reviewed TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP,
    next_review TIMESTAMP WITH TIME ZONE NOT NULL DEFAULT CURRENT_TIMESTAMP,
    times_correct INTEGER NOT NULL DEFAULT 0,
    total_attempts INTEGER NOT NULL DEFAULT 0,
    CONSTRAINT unique_problem_review UNIQUE (problem_id)
);

-- Table: user_streaks
CREATE TABLE user_streaks (
    id UUID PRIMARY KEY DEFAULT uuid_generate_v4(),
    current_streak INTEGER NOT NULL DEFAULT 0 CHECK (current_streak >= 0),
    longest_streak INTEGER NOT NULL DEFAULT 0 CHECK (longest_streak >= 0),
    last_active_date DATE,
    updated_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP
);

Table Properties & Metadata

Table: `problems` (Modified for Google Sync)

Column Name

Data Type

Constraints

Description

id

UUID

Primary Key, Default: uuid_generate_v4()

Unique problem identifier

name

VARCHAR(255)

Not Null, Unique

LeetCode title (e.g., "3Sum")

pattern

VARCHAR(100)

Not Null

Core paradigm (e.g., "Two Pointers")

google_doc_url

TEXT

Nullable

Hyperlink to personal Google Doc notes

google_doc_id

VARCHAR(100)

Unique

Unique identifier of file in Google Drive

reference_code

TEXT

Nullable

The ideal source code extracted from your Doc

5. AI Code Reviewer Integration (Gemini API)

To evaluate handwritten code on a mobile device without compiler dependencies, heavy runtimes, or strict test harnesses, the system integrates the Gemini API using gemini-2.5-flash-preview-09-2025.

API Payload Schema Configuration

import google.generativeai as genai
from google.genai import types

# Define JSON Schema to guarantee structural correctness of the AI response
response_schema = {
    "type": "OBJECT",
    "properties": {
        "is_correct": {"type": "BOOLEAN"},
        "detected_complexity": {
            "type": "OBJECT",
            "properties": {
                "time": {"type": "STRING"},
                "space": {"type": "STRING"}
            },
            "required": ["time", "space"]
        },
        "bugs": {
            "type": "ARRAY",
            "items": {"type": "STRING"}
        },
        "key_suggestion": {"type": "STRING"}
    },
    "required": ["is_correct", "detected_complexity", "bugs", "key_suggestion"]
}

System & Prompt Prompts

system_instruction = """
You are an expert technical interviewer evaluating a candidate's LeetCode code written from memory.
Your primary objective is to review code correctness, algorithmic efficiency, and potential edge-case failures.
Be highly forgiving of minor visual/syntax typos (e.g., off-by-one indentation, minor spelling mistakes in variables, or missing colons)
which would be caught in 1 second by a standard IDE. Focus on raw algorithmic correctness.
"""

user_prompt_template = """
Problem Name: {problem_name}
Target Pattern: {pattern_name}
Expected Complexity: {target_complexities}

Candidate's Code Submission:
{candidate_code}

Evaluate the code logic and return the structured JSON output.
"""

6. System Execution Workflows

Daily Queue Retrieval Workflow

The system minimizes burnout by limiting the daily workload to problems scheduled for on or before the current timestamp.

[User Opens App] 
       |
       v
[Query Database] 
  1. Retrieve Active Streak -> SELECT * FROM user_streaks LIMIT 1;
  2. Lazy Check: If CURRENT_DATE > last_active_date + 1, reset current_streak to 0 in DB.
  3. Query Daily Cards -> SELECT * FROM user_reviews 
                           JOIN problems ON user_reviews.problem_id = problems.id
                           WHERE next_review <= NOW()
                           ORDER BY next_review ASC;
       |
       v
[Render Streamlit Mobile-Responsive Cards & Streak Banner]

Google Drive Ingest Sync (Triggered via UI Button or Cron)

The user can tap "Sync Folder" in the app interface, executing the background sync worker:

                      [User Taps "Sync Google Folder"]
                                     |
                                     v
                        [Read Files from Folder]
             Call Google Drive API: List files inside target FolderID
                                     |
                                     v
                       [Filter New Google Doc IDs]
         Identify Doc IDs not present in problems.google_doc_id DB column
                                     |
                                     v
                  +------------------+------------------+
                  |                                     |
         [No New Docs Found]                    [New Docs Found]
                  |                                     |
                  v                                     v
          Complete Sync Task                    For each new doc:
                                                - Call Google Docs API
                                                - Parse name, patterns, code
                                                - INSERT INTO problems
                                                - INSERT INTO user_reviews 
                                                  (Default Box = 1, next_review = now())
                                                        |
                                                        v
                                                Refresh Review Queue

7. Portability, Accessibility & Performance

1. Fully Responsive Fluid Layouts

Framework Adaptation: All Streamlit elements are configured using native block components (st.columns, st.container) that transition seamlessly from a dual-pane desktop view to a single-stack layout on iOS/Android devices.
Typographic Scaling: Markdown blocks use standard relative heading levels (### and body text) to remain highly readable on varying screen resolutions.
Streak Visual Integration: The app features a high-visibility, fluid CSS indicator displaying the user's current streak count (e.g., "🔥 14 Days Active") pinned to the top of both mobile and desktop screens.
Touch Target Optimization: Touch elements such as "Submit", "Forgot", and "Easy" use high-contrast primary colors with custom internal padding so they can be clicked reliably with a single thumb tap.

2. Manual Offline Fallback Mode (No Connection Option)

Traveling frequently results in spotty network connections. To prevent missed study days and protect your active streak:

The "Self-Grade" Toggle: If connection to the Gemini API is lost or latency is too high, users can toggle a "Manual Grade" override.
Mechanism: The app displays the user's stored Google Doc code and asks the user to manually rate themselves ("Forgot", "Struggled", "Smashed It").
Local Synchronization: The review state transition calculates intervals locally, registers the streak daily increment, and pushes the DB update asynchronously, guaranteeing zero study gaps.

3. Data Protection & Sync Security

Encrypted Cloud Connect: All DB calls are encrypted via TLS/SSL.
Personal Space Partitioning: The Supabase endpoint utilizes Row-Level Security (RLS) to ensure that only you can view, create, or alter your saved algorithmic problem queues and personal study metrics.