DEV Community: Shivani Pal

Echoes : Hacktoberfest Open-Source AI Challenge: Week 4

Shivani Pal — Tue, 27 Oct 2026 15:49:11 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

Between late October and early November—as Halloween shifts into Día de los Muertos—the boundary between what is remembered and what is forgotten feels remarkably thin. We set out ofrendas, light candles, and tell stories of the people whose voices once filled the kitchen. But human memory is fragile. When an elder passes away, an entire library of unwritten family traditions, quiet childhood recollections, and oral histories vanishes with them.

I built Echoes: A Private Memory Keeper.

Echoes is an intimate, local-first web archive designed to capture oral family histories, preserve spoken voices, and catalog personal recollections before they wander away.

Unlike cloud note apps, corporate productivity dashboards, or generic chatbots, Echoes is designed around an unbending core belief: our most personal memories belong to the people who create them, not to a remote cloud server.

Key Highlights

Voice & Oral History Capture: Record spoken memories right through the browser using the native MediaRecorder API, complete with live timing, waveform status, and instant review playback.
Handcrafted Archival Aesthetics:
- Daylight Paper Mode: Warm ivory card stock (#FAF7F2), deep walnut ink, terracotta wax seals, and tactile paper grain.
- Night Study (Candlelight) Mode: Rich dark bookbinding leather (#171513), glowing ember terracotta, eucalyptus sage, and warm candlelight gold.
- Tangible Archival Elements: Accession folio stamps (№ 179311), card catalog book spines, hand-ruled borders, and catalog slips.
100% Local-First Engine: All stories, audio clips, and metadata persist inside the browser's IndexedDB. The application runs 100% offline with zero cloud latency.
Archival Search & Filter: Real-time keyword filtering across stories, people involved, locations, and custom tag pills.
Data Sovereignty: One-click portable JSON export with embedded audio archives, plus permanent erasure options.

Demo

Experience the live deployed application:
👉 https://shivv2430.github.io/Echoes/
Video link : **[ https://drive.google.com/file/d/1_dWSIIfyjWSyiCM3ahGa-GfUKw3Jr3rh/view?usp=sharing]

Tip: Try clicking the "Daylight" / "Night Study" button in the upper right to experience the candlelit archive mode, or click one of the gentle oral history prompts to begin recording.

Code

The complete source code is open source under MIT on GitHub:

shivv2430 / Echoes

https://shivv2430.github.io/Echoes/

Echoes — A Private Memory Keeper

“Memories belong to the people who create them, not to a cloud service.”

Created for the four-week developer challenge themed “Ghost in the Machine.”

🌐 Live Application: https://shivv2430.github.io/Echoes/
📂 GitHub Repository: https://github.com/shivv2430/Echoes

1. Overview

Echoes is an intimate, local-first digital memory keeper designed to help people preserve family oral histories, childhood recollections, cultural traditions, and meaningful moments before they slip away.

Whether recording a grandparent recounting stories from their youth, writing down a treasured family recipe, or documenting a personal milestone, Echoes provides a calm, distraction-free archive.

Unlike typical cloud-based note apps or social media, Echoes operates strictly on your machine. No personal stories, recordings, or metadata are ever transmitted to external cloud databases, analytics services, or proprietary AI servers.

2. Key Features

Calm, Human-Crafted Aesthetic:
- Daylight Paper Mode: Warm ivory card stock (#FAF7F2), deep walnut ink, terracotta wax seals, and subtle…

View on GitHub

GitHub Repository Link: https://github.com/shivv2430/Echoes

How I Built It

Echoes was architected with simplicity, tactile empathy, and complete user sovereignty in mind:

Frontend & Build: Built using React 19 and Vite, maximizing runtime speed and maintaining a lightweight footprint.
Zero-Cloud Storage (IndexedDB + idb): All written memoirs and raw voice recordings are stored directly in the browser's persistent IndexedDB storage engine. Audio recordings are handled as raw binary Blobs, ensuring full playback capability without external hosting.
Audio Capture: Native browser MediaRecorder API with granular microphone permission states, hardware track clean-up, and elapsed timers.
Local-First Open-Weight AI Architecture:
- Echoes is engineered to connect to local open-weight language models via Ollama running on-device (e.g., llama3.2:3b or gemma2:2b).
- Inference executes locally via http://localhost:11434, ensuring that voice transcriptions, memory summarization, and natural language retrieval ("What did Grandpa tell me about his childhood?") never leave the host machine.
- Strict negative constraint prompting prevents the model from hallucinating dates, relationships, or details not found in the archive.

Why Does Open Innovation Matter?

When you record your grandmother talking about fleeing a war, a mother explaining her secret holiday recipe, or your own private confessions from childhood, you are not dealing with "user content"—you are dealing with a person's soul.

If you build this on proprietary cloud APIs:

The Privacy Breach: Private family names, intimate grief, and spoken audio get transmitted to remote data centers, analyzed by closed corporate models, and stored on servers you do not control.
The Obsolescence Trap: When a SaaS startup changes its pricing or shuts down its servers, those memories are locked behind paywalls or erased forever.
The Offline Reality: The moments when oral histories are recorded—around a campfire, in an old country house, or at an elder's bedside—often happen with the lights low and the internet disconnected.

Open innovation changed everything:

Open-weight models (like Llama and Gemma) prove that high-quality synthesis, summarization, and question-answering no longer require sending data across the Pacific Ocean. A compact 3B model running on a laptop's unified memory performs faithfully without a single byte leaving the room.
Open web standards (IndexedDB, MediaRecorder, Web Audio) allow a browser to become an independent archival vault that continues working indefinitely, even if the developer never pushes another update.

Open-source AI transforms technology from a digital landlord into an honest personal instrument.

My Agent Session

This application was designed and built iteratively with Antigravity (Advanced Agentic AI). The pair-programming workflow focused on creating a handcrafted, human-first design system, robust local persistence via IndexedDB, and browser MediaRecorder audio pipeline.

The Spirit of It

During testing, I sat down and recorded a short oral clip of an old family story: the recipe for Sunday vegetable stew that my family made during rainy seasons decades ago, complete with the wooden kitchen table and the smell of cardamom.

Listening back to that recording—sitting in the quiet "Night Study" mode with the warm paper glow and catalog mark № 179311—it didn't feel like staring at software. It felt like holding a real journal.

That is the true ghost in the machine: not a creepy phantom haunting our circuits, but the echoes of the people we love, preserved faithfully on our own terms.

LazyStudy AI : Hacktoberfest Week 3 DEV Challenge

Shivani Pal — Thu, 22 Oct 2026 18:44:24 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built
As a student, I noticed a frustrating pattern during exam season: I was spending hours just organizing, highlighting, and formatting my study material before I ever began actually studying. Manually making flashcards or typing up practice quizzes takes a lot of mental energy, and by the time everything is set up, you are already tired.

I built LazyStudy AI to handle that tedious prep work. It is a lightweight, human-centered study assistant that turns raw lecture notes or textbook passages into four focused study aids:

Smart Summary: Extracts high-impact takeaways, core definitions, and structured overviews without hallucinating details outside the provided notes.
Active Recall Flashcards: Creates atomic question-and-answer pairs with a 3D flip view, keyboard navigation (Space to flip, arrows to navigate), shuffle controls, and mastery tracking.
Practice Quiz: Builds 4-option multiple-choice questions that hide answers until you submit, followed by clear explanations of why the right option holds.
Quick Revision Plan: Splits material into manageable 25 to 35 minute study blocks with an interactive checklist and an explicit reminder that it is an automated suggestion, not an exam guarantee.
Students and self-learners get to be a little lazier with their administrative chores so they can spend their remaining focus on real understanding.

Demo
Live Web App: https://shivv2430.github.io/LazyStudy_AI/
Quick Test: The live demo includes pre-loaded sample notes (Computer Science, Biology, and Psychology) and a built-in simulation mode, so you can test all four tools instantly in your browser without needing anything installed.
Code

shivv2430 / LazyStudy_AI

LazyStudy AI

Your notes. Less effort. More learning.
A lightweight, local-first AI study assistant built with open-weight models for the Hacktoberfest Open-Source AI Challenge (Week 3: Slow and Steady).

🌐 Live Deployment URL: https://shivv2430.github.io/LazyStudy_AI/
(Includes instant Demo Mode simulation so reviewers and students can explore all four study tools directly in their browser without installing anything).

1. Project Overview

Students frequently spend more hours copying, formatting, and reorganizing their study material than actually engaging in active learning. Existing commercial AI tools often demand expensive recurring subscriptions, gatekeep features behind proprietary API quotas, and transmit private student notes—including unpublished dissertations, assignment drafts, and lecture slides—to remote corporate servers.

LazyStudy AI is an open-source, human-centered study companion designed by students, for students. It automates repetitive study preparation by converting unorganized notes into:

Smart Summaries — Crisp overviews, definitions, and hierarchical markdown bullet points.
Active Recall Flashcards — Interactive 3D flip cards with…

View on GitHub

Repository link: https://github.com/shivv2430/LazyStudy_AI

Video Link : https://drive.google.com/file/d/1_8dYBpQfmC3jeChCQYaq09YwcOdrxoDS/view?usp=drive_link

How I Built It
The app is built around open-weight models and local inference:

Local Inference Engine: Ollama running locally on http://127.0.0.1:11434.
Default Model: qwen2.5:3b, an open-weight instruction model that offers reliable structured reasoning and clean JSON output on standard consumer laptops (under 2GB download and 4GB RAM).
Backend Gateway: A Node.js and Express server that handles prompt structuring, JSON schema validation, markdown code fence stripping, and file upload parsing (.txt and .md files up to 250KB). The server communicates with Ollama locally and protects student privacy by never logging private notes.
Frontend: React 18, Vite, Tailwind CSS, and Lucide React icons. I chose a warm academic paper color palette (#FAF8F5) with deep navy text instead of bright neon gradients or generic AI decorative blobs.
Configurability: You can switch models between Qwen 2.5, Gemma 2, or Llama 3.2 by changing a single environment variable (OLLAMA_MODEL) without touching the frontend code.
Why Does Open Innovation Matter?
Students frequently work with sensitive material: unpublished thesis drafts, university lecture slides under strict copyright, or personal essays. Sending all of that raw text to closed cloud AI APIs creates several problems:

Privacy and Data Ownership: When using closed APIs, you have to trust that your notes are not being retained or used for training. With local open-weight models, your raw study notes never leave your personal machine.
Cost and Accessibility: Commercial tools charge recurring monthly subscriptions or bill per token. Students on tight budgets cannot always afford paid subscriptions. Local inference runs on hardware you already own with zero per-request costs.
Longevity and Independence: Proprietary APIs can alter pricing, change system prompts, or deprecate endpoints overnight. Open-weight models like Qwen and Gemma can be downloaded once and run indefinitely, even completely offline without an internet connection.
Open innovation proves that we do not need massive, energy-hungry closed models just to help a student study for a midterm. A well-prompted 3-billion-parameter model running on a laptop is more than capable of doing the job well.

The Lazy Payoff
Before building LazyStudy AI, converting a single 15-page chapter into a set of flashcards and practice questions usually took me around 45 to 60 minutes of manual copying and pasting.

With this tool:

Pasting a lecture note or uploading a markdown outline takes 5 seconds.
A local 3B model produces a complete set of active-recall flashcards and a self-scoring quiz in about 5 to 10 seconds.
I get back roughly 40 minutes of prep time per study session, and I can start testing my memory immediately.

Detour: My Open-Source AI Entry for the Touch Grass Challenge (Challenge Week 1: Touch Grass)

Shivani Pal — Tue, 06 Oct 2026 18:09:34 +0000

**Detour: An AI Agent Designed to Kick You Off Your Screen 🌿

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

I built Detour, an interactive AI web application that flips the script on how we use technology: instead of trying to keep you on the screen, its entire purpose is to kick you off of it.

It is built for anyone who suffers from digital burnout or decision paralysis. Detour guides you through a beautifully animated, soothing 3-step "Vibe Check" questionnaire (asking how much time you have, your current mood, and the environment you want). Based on your specific combination of answers, it generates a personalized real-world mission—like outdoor photography, walking aimlessly, creative writing, or doing a deep self-care reset.

How does it get people into the real world? Once the mission is assigned, the UI completely fades away into a distraction-free "Detour Mode" screen that simply says: "You're good. Put your phone away. I'll be here when you get back." No endless scrolling, no follow-up chats. Just a thoughtful suggestion and a blank screen waiting for your return.

Demo

Live Demo: https://shivv2430.github.io/Detour/

Code

shivv2430 / Detour

Live : https://shivv2430.github.io/Detour/

🌱 Detour

An AI that helps you spend less time with AI.

🚀 Play with the Live Demo Here

The Problem

We have AI assistants that can answer almost anything, generate code, write emails, and summarize entire books in seconds. But in doing so, they also give us another reason to stay glued to our screens. They optimize our digital lives, but often at the expense of our physical ones.

The Idea

Detour is a web application that flips the script. Instead of infinite scrolling, it guides you through a quick 3-step questionnaire to understand how much free time you have, your current mood, and the vibe you want (Nature, Self, Company, etc.). Based on your answers, it generates a personalized real-world mission—like creative writing, walking in nature, cooking, or calling a friend.

And then: It gets out of your way.

The best thing this app can do is know…

View on GitHub

How I Built It

I built the frontend using React, Vite, and Tailwind CSS (v4), heavily utilizing Framer Motion for fluid, glassmorphism animations to make the questionnaire feel calming and premium.

The core logic engine is designed around open-weight models (like Llama 3) via a Node.js/Express backend service layer. The UI takes the user's inputs through the animated questionnaire and translates them into context-rich prompts. This allows the AI model to parse the specific combination of the user's mood and available time to return highly tailored, safe offline missions (e.g., if you are feeling "exhausted" and only have "15 minutes", the open model knows not to suggest a heavy outdoor workout).

Why Does Open Innovation Matter?

Open innovation matters immensely here because an application meant to help you disconnect shouldn't tie you to a closed ecosystem. By designing Detour around open-weight models, we ensure absolute privacy—what you do with your downtime and your current emotional state is your business, not training data for a closed API.

It also makes it possible for anyone to fork the project and run it entirely locally on their own hardware. It gives the power of customization back to the user, proving we don't need a massive, centralized intelligence for something as fundamental as finding a moment of peace in the real world.

New to Hacktoberfest and Already Confused 😅

Shivani Pal — Sun, 04 Oct 2026 18:27:55 +0000

This is my first time participating in Hacktoberfest, so I'm still figuring things out.
I was checking out the Hacktoberfest 2026 challenges on DEV and came across the "Connect your DEV account" option.
I clicked on it, and it takes me to my DEV profile, so I thought my account was connected.
But when I come back to the Hacktoberfest page, it still shows:
"Connect your DEV account"
and the challenges are still locked for me. 😕
So now I'm wondering:
Is my DEV account actually connected, or am I missing a step?
I'm new to Hacktoberfest, so I'm not sure if this is something I did wrong, a browser/session issue, or if the connection is just taking some time to update.
Has anyone else faced this?
If you have participated before or already figured this out, I'd really appreciate some help. 🙌
Am I missing something here, or is anyone else seeing the same thing?

Hacktoberfest #Hacktoberfest2026 #DEVCommunity #OpenSource #WebDevelopment #Beginners #DevChallenges

ContinuEd

Shivani Pal — Sat, 03 Oct 2026 10:40:49 +0000

I Built an AI Content Workflow with Sanity Instead of Just Another Blog 🚀

I wanted to participate in the Sanity challenge, but I didn't want to build another simple blog with a CMS behind it.

So I decided to ask a different question:

What if Sanity wasn't just where I stored content, but where the entire content workflow lived?

That became SanityFlow.

What I built

SanityFlow is an AI-assisted content workflow where content moves through different stages:

📝 Draft

↓

🤖 AI Processing

↓

🔍 Review

↓

👩‍💻 Human Approval

↓

🚀 Published

The idea is simple: an AI agent can help move content forward, but a human still controls the important transitions.

The stack

Next.js
Sanity
Sanity Studio
Sanity App SDK
AI API
TypeScript
React

The interesting part: Sanity is doing more than storing content

My first instinct was to use Sanity as a headless CMS and build the UI separately.

But the challenge pushed me to go deeper.

I experimented with:

1. Custom Sanity interface

Instead of relying completely on the default Studio experience, I created a custom interface for managing the workflow.

The goal was to make it possible to see:

Current content status
AI processing state
Review status
Assigned reviewer
Previous actions

all in one place.

2. Workflow as data

I modeled the workflow alongside the actual content.

That means the content isn't simply:

draft → published

It can have states such as:

draft → processing → review → approved → published

This means an AI agent can work with the same workflow that a human reviewer uses.

3. AI-assisted content

The AI can help with things like:

summarizing content
generating metadata
suggesting tags
identifying missing information
preparing content for review

But instead of letting AI automatically publish everything, the workflow keeps a human approval step.

4. Real-time updates

One of the things I wanted to explore was making the interface feel alive.

When the state of a piece of content changes in Sanity, the custom interface can reflect that change without requiring the user to constantly refresh.

What I learned

The biggest thing I learned from this challenge is that a CMS doesn't have to be just a place where you store documents.

Sanity can become part of the application's logic.

The combination of structured content, custom interfaces, real-time updates and workflows opens up a lot more possibilities than a traditional blog.

What didn't work

Not everything worked on the first try.

I initially approached the project like a normal Next.js application with Sanity attached to it.

That worked, but it wasn't really taking advantage of what Sanity could do.

I also spent more time than expected figuring out how to model workflow states and connect the different pieces together.

There are still parts I'd improve, especially around [YOUR LIMITATION HERE].

But that's probably my favorite part of this build.

I learned more by trying to make the workflow work than I would have by simply following a tutorial.

What's next?

I'd like to take this further by adding:

better AI agents
richer workflow transitions
role-based approvals
analytics
more custom Studio components
automated content transformations

The project is still a work in progress, but that's exactly what I wanted from this challenge.

Build → break → understand → improve.

That's the fun part. 🚀

🔗 Project

GitHub: [https://github.com/shivv2430/ContinuEd---Your-Learning-Always-in-Continuity.]

Live Video Demo: [https://drive.google.com/file/d/1obbbBIG9gbViwLUkYXD-lKinN_Tj2HFc/view?usp=drive_link]

Demo Link : [https://continu-ed-your-learning-always-in.vercel.app/]

If you're also experimenting with Sanity, I'd love to hear what you're building.

BFF-Build For Friend

Shivani Pal — Sat, 03 Oct 2026 09:38:48 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

Most software projects obsess over getting thousands or millions of monthly active users. BFF (Build For a Friend) flips that script completely.

It is a cozy, human-centered project studio built around one single manifesto:

"Pick one real person and build something for them. It doesn't have to be big. It has to matter to them."

Instead of an over-engineered dashboard, BFF showcases three real, functioning micro-tools built for specific loved ones, alongside an interactive builder to plan your own:

🥗 Roommate's Allergy Meal Planner (Built for Alex): Alex has severe peanut and gluten allergies, making shared dinners and takeout a constant source of anxiety. This tool provides an instant allergy safety toggle and a "Surprise Alex" randomizer that guarantees 100% allergy-safe comfort meals in seconds.
🗣️ Patient Practice Partner (Built for Mateo): Mateo is moving to Madrid and gets anxious when ordering coffee or food in Spanish. This tool gives him a zero-judgment conversation buddy with instant gentle coach feedback and pronunciation tips.
🎙️ Grandpa's Voice Memo Recipe Book (Built for Grandpa Joe, Age 84): Grandpa never writes recipes down; he sends rambling 8-minute WhatsApp voice memos. This tool extracts his cooking wisdom (like melting the Parmesan cheese rind into the sauce) into a printable vintage family keepsake card.
✨ The Project Studio: Allows anyone to answer 3 simple questions about a person in their life and immediately save, track, and gift a micro-project to them.

Demo

🎥 ** Video Link:** [https://drive.google.com/file/d/1z0mmmkq0hVumVazXcRdmYnGOQSnplfbv/view?usp=sharing]
🌐 Live Deployed Site: https://shivv2430.github.io/BFF/
📱 Fully responsive on desktop and mobile with zero dependencies or load lag.

Code

shivv2430 / BFF

BFF — Build For a Friend 💛

"Pick one real person and build something for them. It doesn't have to be big. It has to matter to them."

A simple, cozy web app designed to inspire and track micro-projects built with care for an audience of one.

🔗 Live Demo: https://shivv2430.github.io/BFF/
📦 GitHub Repository: https://github.com/shivv2430/BFF

🌟 About The Project

Most coding projects try to reach thousands or millions of users. BFF (Build For a Friend) flips that script:

Audience of One: Build for your roommate, your grandpa, your partner, your sibling, or your best friend.
Micro Over Massive: It can be built in an afternoon or a weekend. A single helpful button or thoughtful feature is better than an overwhelming dashboard.
Gift Culture: Treat software like a handmade birthday card or baked cookies. Send the link in a text message and see their face light up.

…

View on GitHub

GitHub Repository: https://github.com/shivv2430/BFF

How I Built It

I chose to build this with clean, accessible Vanilla HTML5, modern CSS, and plain JavaScript — no massive framework bloat, complex configs, or node_modules black holes.

Local Persistence: Uses localStorage to save your drafted projects directly in your browser.
Interactive Prototyping: Features interactive filters, simulated voice memo transcripts, and conversational dialogue trees.
AI-Assisted Pair Programming: Built iteratively with agentic AI tooling (opencode & agent workflows), focusing on rapid human-first design, empathetic UI/UX, and keeping code accessible and understandable for beginners.

Why Does Open Innovation Matter?

Building for an "Audience of One" embodies the true spirit of open-source and open innovation.

Big tech platforms build for mass commercial scale and engagement metrics. They don't have an incentive to build an app that only one roommate, one grandfather, or one friend will ever use.

Open innovation, open web standards, and accessible AI lower the barrier so that anyone can craft custom software as a personal gift — just like knitting a warm sweater or baking cookies. You don't need venture capital or closed proprietary APIs to build something that brings tears of joy to someone you care about.

My Agent Session

This project was planned, designed, and iterated using agentic pair programming via the terminal and developer environment:

Iterative scaffolding of the 3 starter concepts into working interactive prototypes.
Human-in-the-loop feedback to deliberately simplify the design, ensuring it feels warm, authentic, and approachable rather than an impersonal enterprise SaaS template.

Prize Categories

Build for a Friend (Grand Prize & Community Favorite)
Most Heartwarming / Audience of One
Clean & Accessible Web -Created by Shivani Pal

Hello DEV 👋 I’m Shivani, and I’m Building My Way Into Tech

Shivani Pal — Sat, 03 Oct 2026 08:14:12 +0000

**Hello DEV 👋 I’m Shivani, and I’m Building My Way Into Tech

Hey everyone! 👋**

I’m Shivani, an engineering student who is slowly figuring out my way around the world of tech.

I started with the basics of HTML and CSS, then moved into JavaScript, Git & GitHub, and eventually started exploring React, APIs, and backend development.

And honestly, it hasn’t always been smooth. 😅

There have been plenty of:

“Why is this code not working?” moments
Git errors I didn’t understand
Deployments that worked locally but broke online
Projects that looked much better in my head 😂

But every broken project has taught me something.

Recently, I’ve also been exploring open source, hackathons, DSA, and building projects instead of just watching tutorials.

I’m joining DEV because I want to document that journey.

Here, I’ll be sharing:

💻 Things I learn while coding

🛠️ Projects I build

🐛 Bugs I somehow manage to create

📚 Resources that actually help me

🌱 My experiences with open source and hackathons

🚀 And the lessons I learn along the way

I’m still learning, still experimenting, and definitely still Googling things.

But I’m excited to see where this journey goes.

If you're also learning to code, feel free to say hi! 👋

Let’s learn and build together. 💙