DEV Community: Aditya Pathak

LocalBiz AI

Aditya Pathak — Wed, 14 Oct 2026 09:35:22 +0000

LocalBiz AI

LocalBiz AI is a focused marketing workspace for small businesses, local shops, cafés, makers, and online sellers. It turns product details, promotions, and everyday business ideas into useful marketing content without adding a heavy content-management system.

Features

Public landing page with Product Writer, Social Studio, and Language Assistant previews
Supabase email/password authentication with SSR session cookies
Protected dashboard and server-authenticated /api/generate
Groq chat completions using the configurable open-weight model openai/gpt-oss-20b
Product copy, social posts, and English/Hindi marketing translation
Browser-only drafts and business preferences, namespaced per authenticated user
Copy, regenerate, save, reopen, and delete draft actions
Responsive dashboard, auth screens, loading states, validation, and error messages

Local development

npm install
cp .env.example .env.local
npm run dev

Open https://www.localblizai.dev/.

Configure Supabase Auth

Create a Supabase project, enable Authentication → Providers → Email, and copy the project URL and publishable/anon key into .env.local:

NEXT_PUBLIC_SUPABASE_URL=https://your-project.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_publishable_or_anon_key

The app uses @supabase/ssr with createBrowserClient, createServerClient, and middleware cookie refresh. It does not store passwords, session tokens, or user IDs in localStorage. If Supabase variables are missing, the public homepage still renders but auth and the private dashboard show a clear configuration message; the Groq route returns a configuration error rather than bypassing authentication.

In Supabase Authentication → URL Configuration, set:

Site URL (local): http://localhost:3000
Redirect URL (local): http://localhost:3000/auth/callback
Redirect URL for password reset: http://localhost:3000/auth/callback?next=/reset-password

For production, replace these with your deployed Vercel URL, for example:

https://your-domain.vercel.app
https://your-domain.vercel.app/auth/callback
https://your-domain.vercel.app/auth/callback?next=/reset-password

Configure Groq

Add the server-only Groq credentials:

GROQ_API_KEY=your_key_here
GROQ_MODEL=openai/gpt-oss-20b

GROQ_API_KEY must not use the NEXT_PUBLIC_ prefix. The key is only read in the server route after the authenticated user is verified. Requests are validated and limited to 2,000 characters per field. Missing keys, rate limits, upstream failures, and malformed model responses return useful errors. Inference is hosted by Groq; it is not local inference.

Vercel deployment

Import the repository into Vercel.
Add NEXT_PUBLIC_SUPABASE_URL, NEXT_PUBLIC_SUPABASE_ANON_KEY, GROQ_API_KEY, and optionally GROQ_MODEL to the Vercel project environment variables.
Deploy the project.
Update Supabase Site URL and Redirect URLs to the final production URL.
Test signup confirmation, sign-in, sign-out, password reset, dashboard generation, and a second user’s draft isolation.

Why open-weight AI

Open-weight models give LocalBiz AI a more portable generation layer than a provider-locked workflow. Today the app uses fast hosted Groq inference; later, a compatible open-weight deployment can be introduced without changing the product’s core prompts and UI. This is useful for small businesses that need practical content creation while keeping options open around cost, hosting, and model choice.

TrailLens

Aditya Pathak — Wed, 07 Oct 2026 16:21:11 +0000

🌿 TrailLens — Offline AI Nature Explorer

TrailLens is an outdoor-first AI nature companion built for the Hacktoberfest Open-Source AI Challenge, Week 1 — Touch Grass.

Take a photo of something you find outside. TrailLens identifies it with Ollama + Qwen3-VL 2B, explains it in simple language, gives you three facts, and gives you an outdoor challenge.

The screen is not the destination. The screen gives you your next reason to go outside.

Why open innovation matters

The core image understanding runs locally through Ollama and the open-weight Qwen3-VL 2B vision model. The local experience can analyze photos without sending them to a closed AI provider, and the model can be swapped or changed by the developer.

Features

📷 Camera-friendly photo upload
🧠 Local image understanding with Ollama + Qwen3-VL 2B
🌿 Beginner-friendly identification and facts
🎯 Outdoor challenges designed to get people exploring
⚡ XP and levels
📓 SQLite discovery journal
🌐 Explicit Demo Mode for hosted deployments

Architecture

Photo
  ↓
FastAPI
  ↓
Ollama
  ↓
Qwen3-VL 2B
  ↓
Identification + Facts + Outdoor Challenge
  ↓
SQLite Journal + XP

Local setup

Requirements:

Python 3.10+
Ollama

Pull the model:

ollama pull qwen3-vl:2b

Install dependencies:

python -m pip install -r requirements.txt

Run:

python -m uvicorn main:app --host 127.0.0.1 --port 8000

Open https://traillens-26.onrender.com.

Local AI vs online demo

Local Mode is the core experience:

![ ](https://dev-to-uploads.s3.us-east-2.amazonaws.com/uploads/articles/wk60gegrwa58fbllakhs.png)
![ ](https://dev-to-uploads.s3.us-east-2.amazonaws.com/uploads/articles/6d7m5elf1w64fmn7wk03.png)
Your Mac
  ↓
FastAPI
  ↓
Ollama
  ↓
Qwen3-VL 2B

No cloud AI is required for image analysis.

A hosted Render service cannot access the Ollama process running on your Mac. For that reason, the included Render configuration enables an explicitly labelled Demo Mode instead of pretending that local Ollama is running remotely.

Render deployment

The repository includes render.yaml.

Build command:

pip install -r requirements.txt

Start command:

uvicorn main:app --host 0.0.0.0 --port $PORT

The hosted service uses Demo Mode. The real local open-weight AI remains available when TrailLens is run locally with Ollama.

Repository structure

trail-lens/
├── main.py
├── backend/
│   ├── __init__.py
│   ├── config.py
│   ├── database.py
│   ├── demo.py
│   ├── ollama.py
│   └── xp.py
├── frontend/
│   ├── index.html
│   ├── journal.html
│   ├── app.js
│   ├── journal.js
│   └── styles.css
├── data/
├── uploads/
├── .env.example
├── .gitignore
├── .python-version
├── render.yaml
├── requirements.txt
└── README.md

License

MIT.

Study Buddy

Aditya Pathak — Fri, 02 Oct 2026 16:32:58 +0000

What I Built

I built StudyBuddy for my close friend Rohan, who is currently juggling a grueling preparation schedule for software engineering technical interviews and system design certifications while working a demanding job.

The Problem Rohan Faced

Whenever Rohan sat down to study in the evenings, he hit three recurring roadblocks:

Focus Fragmentation & Tab Drifting: Traditional web timers drift by several seconds or stall entirely when backgrounded in another tab during coding exercises, ruining pomodoro rhythms.
The "Illusion of Competence": Passively rereading notes or documentation gave him a false sense of mastery, only for the knowledge to evaporate days later (the classic Ebbinghaus Forgetting Curve).
App Bloat & Paywalls: Popular flashcard and quiz apps are clunky, loaded with subscription paywalls, require mandatory account sign-ups, or bombard the user with ads right when deep focus is needed most.

The Solution: StudyBuddy

StudyBuddy is a zero-bloat, distraction-free study companion that unites the three most scientifically backed learning methods into one unified, offline-first web experience:

⏱️ Zero-Drift Pomodoro Timer: Engineered with monotonic delta timing and tab visibility recalibration, paired with native Web Audio chime alerts and a soothing synthesized Lo-Fi ambient rain generator.
🗂️ Interactive 3D Flip Flashcards: Tactile 3D perspective flip cards with 4-level Leitner-inspired recall feedback (Again, Hard, Good, Easy) that dynamically calculates deck mastery.
📝 Smart Quiz Engine: Instantly transforms any flashcard deck into an interactive 4-option multiple choice test or rapid recall drill with automated distractor generation, countdown timers, and celebration confetti.
🎨 Deep Customization: 4 curated themes (Midnight Nebula, Daylight Paper, Forest Calm, Sunset Glow), custom timer durations, and full JSON import/export so Rohan owns all his study notes with zero cloud lock-in.

Demo

StudyBuddy runs instantly in any modern web browser with zero build steps or external dependencies.

Live Local Demo: https://aditya50032.github.io/Study-Partner/
Zero Installation: Simply open index.html in your browser.

Key Interactive Highlights

Pomodoro Dial: Glowing circular SVG progress ring, session cycle tracker (e.g. 🍅 1 of 4), and quick preset chips (15m, 25m, 45m, 50m).
3D Flip Interaction: Press Space or click to flip the card in 3D space with subtle ambient lighting.
Smart Quiz: Instant feedback highlighting correct answers in emerald green and mistakes in ruby red with full explanations.
Sound Design: Pure Web Audio synthesis — dual-harmonic sine bells (528Hz Solfeggio + 660Hz) and brown-noise rain generator.

Code

Aditya50032 / StudyBuddy

What I Built

The Problem Rohan Faced

Whenever Rohan sat down to study in the evenings, he hit three recurring roadblocks:

Focus Fragmentation & Tab Drifting: Traditional web timers drift by several seconds or stall entirely when backgrounded in another tab during coding exercises, ruining pomodoro rhythms.
The "Illusion of Competence": Passively rereading notes or documentation gave him a false sense of mastery, only for the knowledge to evaporate days later (the classic Ebbinghaus Forgetting Curve).
App Bloat & Paywalls: Popular flashcard and quiz apps are clunky, loaded with subscription paywalls, require mandatory account sign-ups, or bombard the user with ads right when deep focus is needed most.

The Solution: StudyBuddy

StudyBuddy is a zero-bloat…

View on GitHub

🔗 GitHub Repository: github.com/Aditya50032/MyStudyBuddy

The entire codebase is open-source, standalone, and built with pure HTML5, Vanilla CSS, and modern JavaScript:

DEV project/
├── index.html       # Semantic layout, SVG dials, accessible modals & tab controls
├── styles.css       # 4 theme palettes, 3D card transforms, glassmorphic surfaces
└── app.js           # Zero-drift timer, Web Audio synthesis, Leitner rating & quiz engine

Core Architecture Highlights

1. High-Precision Zero-Drift Clock Math (`app.js`)

Standard setInterval(counter--, 1000) drifts severely due to main-thread event loop lag and background tab throttling. StudyBuddy derives remaining time from monotonic wall-clock deltas and hooks into the visibilitychange event:

recalibrate() {
  if (!this.isRunning) return;
  const now = performance.now();
  this.remainingMs = Math.max(0, this.targetEndTime - now);
  if (this.remainingMs <= 0) {
    this.finishSession();
  } else {
    this.onTick(this.getState());
  }
}

2. Asset-Free Native Audio Synthesis

Rather than loading heavy MP3 files that fail or get blocked by browser autoplay policies, sounds are synthesized dynamically:

playTimerChime() {
  const now = this.ctx.currentTime;
  const osc1 = this.ctx.createOscillator();
  const osc2 = this.ctx.createOscillator();
  const gain = this.ctx.createGain();

  osc1.type = "sine";
  osc1.frequency.setValueAtTime(528, now); // Solfeggio frequency
  osc2.type = "sine";
  osc2.frequency.setValueAtTime(660, now + 0.1);

  gain.gain.setValueAtTime(0.3, now);
  gain.gain.exponentialRampToValueAtTime(0.0001, now + 2.2);

  osc1.connect(gain);
  osc2.connect(gain);
  gain.connect(this.ctx.destination);
  osc1.start(now);
  osc2.start(now + 0.1);
  osc1.stop(now + 2.2);
  osc2.stop(now + 2.2);
}

3. Automated Distractor Generation for Quizzes

When a quiz starts, the engine extracts the correct answers and dynamically synthesizes plausible distractors from the remaining deck pool to challenge active recall.

How I Built It

I built StudyBuddy through an interactive agentic pair-programming workflow using Google Antigravity and Gemini 3.8 Flash, grounding architectural decisions in proven open community patterns:

Community Research Pass: Before writing a single line of code, we queried developer wisdom on dev.to to address known pitfalls with web timers. RayLabs' article, Build a Zero-Drift Web Timer, provided the foundation for our monotonic clock delta math and Page Visibility recalibration.
Cognitive Science Principles: We pre-loaded starter decks covering the Ebbinghaus Forgetting Curve, Leitner Spaced Repetition, and Active Recall so that users immediately experience why testing beats rereading.
Local-First & Resilient: We avoided heavy frontend dependencies (like React or Webpack) in favor of zero-dependency modern vanilla web standards, ensuring instantaneous 60fps load times and absolute portability.

Why Does Open Innovation Matter?

Open innovation is what made building StudyBuddy possible in a single weekend:

Complete Privacy & Zero Data Harvesting: Rohan's study notes, interview questions, and focus logs belong strictly to him. In an era where educational platforms monetize user data and lock essential features behind paywalls, an open-source, local-first app guarantees 100% data sovereignty.
Works Everywhere, Even Offline: Open web standards mean this app runs without internet connectivity on a subway commute, in a quiet library corner, or on an airplane.
Extensibility for the Community: Because the code is open and cleanly decoupled, anyone can fork it, contribute new flashcard packs, add LaTeX math formatting, or integrate local open-weight AI models (like Gemma 2 via Ollama) to automatically generate flashcards directly from lecture PDFs and raw notes.

My Agent Session

This entire application was designed, researched, implemented, and verified in an agentic coding session with DevRelay. You can inspect the full prompt trajectory, research calls, and implementation steps below:

Building StudyBuddy: Zero-Drift Pomodoro, 3D Flashcards & Quiz Engine

You

build a study buddy app with pomodoro clock and flash card and quiz where user can set it make it simple app

Agent

I will build StudyBuddy, an all-in-one focus and active recall companion featuring a zero-drift Pomodoro timer, 3D flip flashcard decks, and an interactive quiz generator with local data persistence and custom themes. Let's research community best practices first.

▸ search_dev_to_semantic query: building a pomodoro timer in javascript drift Date.now web audio

Input

query: building a pomodoro timer in javascript drift Date.now web audio

Output

Found article: Build a Zero-Drift Web Timer by raylabs

▸ write_to_file TargetFile: index.html

Input

TargetFile: index.html

Output

Created index.html with Pomodoro dial, 3D flashcards, quiz layout, and deck manager

▸ write_to_file TargetFile: styles.css

Input

TargetFile: styles.css

Output

Created styles.css with 4 themes, 3D card perspective, and glassmorphism

▸ write_to_file TargetFile: app.js

Input

TargetFile: app.js

Output

Created app.js with zero-drift timer, Web Audio API synthesis, flashcard Leitner ratings, and quiz engine

Application built, verified with node syntax check, and serving locally at http://localhost:3000

🔗 Direct Session Link: View Agent Session on DEV

Prize Categories

Hacktoberfest Weekend Challenge: Build for a Friend (Primary Entry)
Best Use of Open Tools & Local AI Architecture

Built with ❤️ for Rohan and lifelong learners everywhere.

DEV Community: Aditya Pathak

LocalBiz AI

LocalBiz AI

Features

Local development

Configure Supabase Auth

Configure Groq

Vercel deployment

Why open-weight AI

TrailLens

🌿 TrailLens — Offline AI Nature Explorer

Why open innovation matters

Features

Architecture

Local setup

Local AI vs online demo

Render deployment

Repository structure

License

Study Buddy

What I Built

The Problem Rohan Faced

The Solution: StudyBuddy

Demo

Key Interactive Highlights

Code

Aditya50032 / StudyBuddy

What I Built

The Problem Rohan Faced

The Solution: StudyBuddy

Core Architecture Highlights

1. High-Precision Zero-Drift Clock Math (app.js)

2. Asset-Free Native Audio Synthesis

3. Automated Distractor Generation for Quizzes

How I Built It

Why Does Open Innovation Matter?

My Agent Session

Prize Categories

1. High-Precision Zero-Drift Clock Math (`app.js`)