DEV Community: Onkar Bhattacharya

# 🌿 Gentle Companion: A Compassionate Path to Mental Well-being

Onkar Bhattacharya — Tue, 05 May 2026 13:11:59 +0000

Introduction

In a world that often demands constant productivity and relentless positivity, it's easy to feel overwhelmed by the weight of depression. Many digital tools promise solutions, but few offer true understanding and gentle support. Enter Gentle Companion, a web app designed to be a safe, evidence-based, and deeply compassionate ally in your journey towards mental well-being. It aims to make therapeutic skills, self-tracking, and micro-actions accessible to everyone, every day, without judgment or pressure.

My Vision: No One Faces Depression Alone

Gentle Companion was born from a singular, powerful principle: any action is worthy of celebration. My vision is a world where every person has a gentle, effective, private companion that helps them reclaim hope, one tiny step at a time. The app never scolds, never punishes missed days, and never demands more than you can give. It's about meeting you where you are, with kindness and understanding.

Thoughtful Features for Gentle Progress

Gentle Companion offers a suite of carefully crafted features, all designed to support you without overwhelming you. Here are some of the key functionalities:
| Feature | Description |
| - -| - -|
| Mood & Energy Check-in | A daily slider for energy/weight (1–10) and a "mood weather" selector (stormy → sunny). Skippable with a "Just let me in" button, ensuring no pressure. |
| Daily Micro-Plan | Three randomly selected tiny tasks from a curated library. You can swap any task for a different one. Completion triggers a gentle celebration, focusing on progress, not perfection. |
| Thought Untangler | A 4-step guided Cognitive Behavioral Therapy (CBT) reframing tool. Sessions are persisted, allowing you to revisit past reflections. |
| Glimmer Catcher | A space to log anything even slightly okay - one sentence is enough. These small moments of positivity are displayed in reverse-chronological order. |
| Soothing Toolkit | Includes a box breathing pacer (4–4–4–4 animated circle), 5–4–3–2–1 senses grounding exercise, and a safe place visualization. |
| Crisis Button | An always-visible "I need help" button that provides immediate access to crisis lines (988, Samaritans, IASP), SMS options, and a pre-written message to a trusted contact. |
| Adaptive Quiet Mode | When your logged energy is ≤ 3, the home screen automatically switches to a minimal layout with only two gentle options, surfacing your "Letter to Myself" if one exists. |
| Letter to Myself | Write a compassionate note to yourself on a stable day, which is automatically surfaced on low-energy days for gentle encouragement. |
| Weekly Insights | Passive mood bar charts, glimmer correlation insights (e.g., "on days you caught a glimmer, your energy was higher"), and a virtual garden that grows with your engagement. |
| Virtual Garden | One element grows per week of engagement, never lost on missed days. A wildflower for the messy weeks, symbolizing growth even through challenges. |

Privacy-First Design: Your Data, Your Control

In an age where data privacy is paramount, Gentle Companion stands out with its unwavering commitment to user privacy. This app is designed with privacy as a first-class feature, not an afterthought. Here's how:

No server, no network requests: All your data is stored exclusively in your browser's local storage. Nothing is transmitted anywhere, ensuring complete confidentiality.
Consent before storage: No data is read from or written to local storage until you explicitly give consent on the privacy screen.
Right to erasure: A two-step "Delete everything" option in the settings allows you to remove all your data and withdraw consent with a single tap.
No third-party data sharing: There are no analytics, error tracking, advertising SDKs, or external CDNs (like Google Fonts) that could potentially leak your IP address or data.
Content Security Policy: A robust Content-Security-Policy meta tag restricts all sources to 'self', enhancing security. ## Grounded in Evidence: Our Therapeutic Framework Every feature and interaction within Gentle Companion is grounded in evidence-based therapeutic approaches, ensuring that the support you receive is both gentle and effective: | Framework | Where Used in Gentle Companion | | - -| - -| | Behavioural Activation (BA) | Daily Micro-Plan: Encourages scheduling small, rewarding activities. | | Cognitive Behavioural Therapy (CBT) | Thought Untangler: Guides you in identifying and gently challenging unhelpful thought patterns. | | Self-Compassion (Kristin Neff) | Letter to Myself, all app copy tone, and celebration language: Fosters self-kindness and understanding. | | Acceptance & Commitment Therapy (ACT) | Glimmer Catcher: Promotes noticing experiences without judgment; also integrated into Thought Untangler. | | Polyvagal-inspired grounding | Box Breathing, 5–4–3–2–1 Senses, Safe Place Visualisation: Techniques to regulate the nervous system. | | Self-Determination Theory | No streaks, no leagues: Emphasizes autonomy (choosing tasks), competence (celebrating completion), and relatedness (virtual garden). | Important Note: Gentle Companion is a supportive companion, not a medical device or a substitute for professional mental health care. The crisis button is always visible and connects users to real human support when needed. ## Tech Stack Highlights For those interested in the technical underpinnings, Gentle Companion is built with a modern, dependency-light stack:
UI Framework: React 18.2
Language: TypeScript 5.3
Routing: React Router DOM 6.20
Styling: Pure CSS Custom Properties (no CSS-in-JS library)
State Management: React Context + useState (no external state library)
Persistence: Browser localStorage Notably, the project boasts zero runtime dependencies beyond React, React Router, and Lucide, emphasizing performance and maintainability. ## Trying out GitHub - OnkarBhattacharya/gentle-companion Contribute to OnkarBhattacharya/gentle-companion development by creating an account on GitHub.github.com Gentle Companion A compassionate path to mental wellbeing. gentlecompanion.netlify.app

Join the Journey

Gentle Companion is more than just an app; it's a philosophy of kindness and support. We invite you to explore this compassionate tool and join us in building a world where mental well-being is accessible to all, one gentle step at a time.
- -
Author: Onkar Bhattacharya
License: MIT © Gentle Companion Contributors
"You don't have to fix anything right now. Just being here is enough."

AgroAdvisor

Onkar Bhattacharya — Fri, 24 Apr 2026 14:21:20 +0000

Empowering Farmers with AI: Building AgroAdvisor — A Multi-Model Advisory Platform

In many parts of the world, agriculture isn't just an industry; it's a lifeline. Yet, farmers often face devastating losses due to crop diseases, unpredictable yields, and pest infestations. What if they had a specialized agricultural expert in their pocket, available 24/7 in their local language?

Enter AgroAdvisor, an intelligent agricultural advisory system designed to bridge the gap between advanced machine learning and field-level implementation.

The Vision: AI That Listens and Learns

AgroAdvisor isn't just another chat app. It’s a comprehensive ecosystem that combines real-time messaging, multilingual translation, and a suite of specialized ML models to provide actionable advice.

The 4-Phase Build Strategy

We built AgroAdvisor using a phased approach to ensure stability and rapid deployment:

The API Layer: A thin routing layer using keyword-based intent classification.
The Core Models: Training specialized models for disease detection, yield prediction, and pest risk.
Accessibility: Breaking the language barrier and adding human expert escalation.
The Flywheel: Creating a continuous retraining loop based on real-world feedback.

Under the Hood: The Multi-Model Engine

At the heart of AgroAdvisor are three distinct models, each optimized for a specific agricultural challenge:

1. Crop Disease Detection (Computer Vision)

Using a ResNet-50 architecture fine-tuned on the PlantVillage dataset (~54,000 images), the system can identify 38 different classes of crop diseases from a simple photo. With image augmentation and transfer learning, we achieved a validation accuracy of over 95%.

2. Yield Prediction (Tabular Regression)

Farmers need to plan for the future. We trained an XGBoost regressor on a combined dataset of NASA POWER weather history and FAO GAEZ soil quality metrics. By analyzing rainfall, temperature, and soil type, AgroAdvisor provides precise harvest estimates.

3. Pest Risk Scoring (Interpretable AI)

Pest management is about timing. We used a Decision Tree classifier trained on iNaturalist occurrence data. By keeping the model interpretable, we ensure that human experts can easily verify the "rules" the AI is using to score risk levels as Low, Medium, or High.

Breaking Barriers: Language and Trust

AI is only useful if it's accessible. AgroAdvisor addresses two major hurdles in agricultural tech:

🌍 The Language Barrier

Farmers often speak local dialects. We integrated a Translation Service supporting Google Translate and LibreTranslate, ensuring that advice delivered in English can be instantly translated into Swahili, Hindi, or any other local language.

🤝 The Trust Gap

No AI is perfect. To ensure reliability, we implemented a 70% Confidence Threshold. If the AI is unsure of its answer, it doesn't guess. Instead, it flags the message and automatically routes the query to a human expert via the existing group chat feature.

The Data Flywheel: Continuous Improvement

The most powerful feature of AgroAdvisor is its ability to get smarter over time. Every response includes a "Helpful" or "Wrong" feedback button.

When a farmer provides a label, it is instantly stored in a Retraining Dataset in Firestore. Our automated pipeline can then export this data to fine-tune the models, ensuring the AI adapts to real-world conditions and regional variations.

Tech Stack

Backend: Flask & Flask-SocketIO
ML Frameworks: PyTorch (ResNet), XGBoost, Scikit-Learn
Database: Google Cloud Firestore
Deployment: Docker & Cloud Run
Translation: Google Cloud Translation API / LibreTranslate

Conclusion: Scaling the Impact

AgroAdvisor represents a shift toward more practical, human-centric AI. By combining specialized models with a feedback-driven retraining loop and multilingual accessibility, we are building a tool that doesn't just predict the future—it helps farmers grow it.

Check out the project on GitHub:

Have thoughts on AI in agriculture? Let’s discuss in the comments!

PortfolioForge: Building an AI‑Native Portfolio Platform with Next.js, Firebase, and Genkit

Onkar Bhattacharya — Sun, 19 Apr 2026 18:10:39 +0000

Github : https://github.com/OnkarBhattacharya/PortfolioForge

Test link : https://studio--studio-3849653404-e5627.us-central1.hosted.app/

PortfolioForge: Building an AI‑Native Portfolio Platform with Next.js, Firebase, and Genkit

Creating a professional portfolio shouldn’t feel like a full‑time job. Yet for many developers, designers, and digital professionals, it often does. Between gathering work, writing descriptions, designing layouts, and deploying a site, the process becomes a barrier rather than an enabler.

I wanted to fix that.

PortfolioForge is an open‑source, AI‑powered platform I built to make portfolio creation fast, intelligent, and accessible. It’s designed as a modern, production‑grade application that uses Generative AI not as a gimmick, but as a core engine for real productivity.

This post breaks down the why, the how, and the tech behind the project.

🚀 The Problem: Portfolio Creation Is Too Hard

Even highly skilled professionals struggle with at least one of these:

Design — making something visually appealing
Development — building and deploying a site
Storytelling — writing compelling project summaries
Time — gathering work and formatting everything

A portfolio is essential, but the process is painful.

PortfolioForge flips that experience by acting as a co‑pilot for your professional identity.

🤖 The Solution: An AI‑Powered Portfolio Co‑Pilot

PortfolioForge automates the hardest parts of portfolio creation using three core capabilities.

1. Multi‑Modal Data Ingestion

Users can upload or link almost anything:

PDF or image‑based CVs
Raw LinkedIn profile text
Public GitHub repositories
Any public URL (blogs, project pages, case studies)

The system parses, structures, and extracts meaningful data automatically.

2. AI‑Generated Content & Summaries

Using Google Genkit, the platform can:

Generate headlines and summaries
Rewrite or refine descriptions
Summarize long‑form content
Turn a GitHub repo into a polished portfolio entry

It’s like having an editor who understands your work.

3. Generative Theme Design

Instead of rigid templates, users describe a theme:

“A calming ocean breeze”

“A futuristic neon aesthetic”

“Warm earthy tones with minimalist typography”

The system generates a unique color palette and visual identity on the fly.

🏗️ Technical Architecture

PortfolioForge is built with a modern, scalable, and secure stack designed for real‑world production use.

Frontend: Next.js 15 (App Router)

Why Next.js?

Server‑side rendering for performance
Strong SEO out of the box
Smooth routing and data fetching
Excellent DX for building modern web apps

The UI is fully componentized and optimized for fast rendering.

Backend: Firebase (Serverless Architecture)

Firebase powers the entire backend:

Firestore — NoSQL database with strict, user‑ownership‑based security rules
Firebase Authentication — Google, Apple, and anonymous sign‑in
Firebase Hosting — Global CDN, auto‑scaling, zero‑maintenance
Firebase App Check — Protects backend from unauthorized clients

This setup eliminates server management and scales automatically.

AI Engine: Google Genkit

Genkit handles all generative features:

Content generation
Summaries
Theme creation
Narrative refinement

A key architectural decision:

AI models are dynamically managed using Firebase Remote Config, allowing model switching (e.g., gemini‑1.5‑flash) without redeploying the app.

This makes the system future‑proof.

🔐 Security & Production Readiness

PortfolioForge isn’t a prototype — it’s engineered for production.

Environment‑based credential management
Granular Firestore & Storage security rules
CI/CD pipelines for consistent deployments
Structured logging
Core Web Vitals monitoring
Full production audit for security and scalability

Everything is built with real‑world reliability in mind.

🌍 Impact on the Developer Ecosystem

PortfolioForge contributes to the digital tech community in three ways:

1. Empowering Professionals

Anyone can create a polished portfolio without needing:

Design skills
Web development expertise
Hours of manual formatting

2. Open‑Source Learning Resource

The codebase demonstrates:

How to architect an AI‑native application
Best practices for Firebase security
Real‑world Genkit integration
Scalable Next.js 15 patterns

3. Moving AI Beyond Chatbots

PortfolioForge shows how AI can be embedded deeply into a product to solve practical, user‑centric problems.

🎯 Final Thoughts

PortfolioForge represents my vision for the future of professional identity: intelligent, automated, and beautifully personalized.

As the sole architect and developer, I built this platform to push the boundaries of what AI‑native applications can look like — combining engineering, design, and generative intelligence into a cohesive product.

And this is just the beginning.

Building an Interactive Wind Turbine Calculator

Onkar Bhattacharya — Sun, 19 Apr 2026 18:09:10 +0000

Small-scale wind energy has always faced a challenge: accurately estimating how much power a turbine will produce at a specific location. While solar energy benefits from numerous calculators, wind energy tools often fall short—either too simplistic or too technical for everyday users.

This post introduces a modern, interactive wind turbine calculator designed to bridge that gap. It leverages real NASA wind data, authentic turbine power curves, and engineering models to provide homeowners, farmers, and small businesses with clear insights into the feasibility of installing 1–10 kW turbines at their sites.

Why Build a Wind Turbine Calculator?

Small wind energy is a niche but growing sector, vital for rural electrification, off-grid cabins, microgrids, and hybrid solar-wind systems. However, wind resources vary dramatically by location, making reliable production estimates essential.

The calculator taps into NASA POWER's global wind speed and direction data, applying advanced engineering models to estimate:

Annual energy production

Monthly and seasonal variations

Number of turbines needed to meet energy targets

Land area and turbine spacing requirements

Losses including wake effects, electrical inefficiencies, icing, and degradation

All calculations happen live in the browser, offering instant feedback.

How the Calculator Works

At its core, the tool answers: "How many small wind turbines do I need to generate X kWh per year at my location?"

Fetch NASA POWER Wind Climatology

Multi-year averages

Wind speed at 10 meters

Wind direction distribution

Apply Wind Engineering Models

Weibull distribution for wind probability

Power law to adjust wind speed to hub height

Air density correction based on elevation

IEC-61400 and IEA loss models

Simulate Real Turbine Behavior

Includes models for 1 kW, 3 kW, 5 kW, and 10 kW turbines, detailing rotor diameter, cut-in, rated, and cut-out speeds, and full power curves.

Calculate Annual Energy Production (AEP)

Integrates turbine power curves with Weibull distributions for realistic output estimates.

Visualize the Wind Rose

Generates an animated directional wind rose using a von Mises distribution, rendered in SVG/Canvas.

Compute Spacing and Land Requirements

Based on industry norms:

5–9× rotor diameter downwind

3–5× crosswind

User Interface

The calculator features a modern SaaS-style two-panel layout:

Left panel: Sticky sidebar with input controls

Right panel: Live-updating results

Built with TypeScript, React + Vite, TailwindCSS, and Radix UI, it offers a fast, clean, and responsive experience.

Project Structure

client/: React app, UI, and wind calculation engine

server/: Express server for production builds

shared/: Shared constants and logic

Key files include windCalculations.ts (AEP, losses, Weibull, spacing) and turbineModels.ts (power curves and turbine specs).

Example: Estimating Wind Output in London

Input:

Field

Value

Latitude

51.5074

Longitude

-0.1278

Target Energy

100,000 kWh

Turbine Size

10 kW

Terrain

Suburban

The calculator fetches NASA wind data, estimates hub-height wind speed, applies losses, simulates the turbine, and outputs the number of turbines needed, wind rose visualization, and land area estimates—all in real time.

Customization and Extensibility

The project is open and hackable:

Add new turbines by editing turbineModels.ts

Adjust loss assumptions in windCalculations.ts

Modify UI themes via Tailwind and CSS variables

Swap in different wind datasets if needed

Ideal for off-grid system designers, renewable energy students, microgrid planners, and DIY wind enthusiasts.

Contributing

Contributions are welcome, especially for:

New turbine models

Improved loss modeling

UI/UX enhancements

Additional visualizations

The repository uses a simple branching model:

main: stable

feature/*: active development

References & Credits

This project builds on work from:

NASA POWER Project

IEC 61400-2 small wind standards

IEA Wind Task 11

Open-source wind modeling research

Final Thoughts

Small wind energy is often misunderstood but holds great potential when analyzed with accurate data and models. This calculator aims to make wind energy analysis accessible, transparent, and enjoyable.

Explore the open-source code, contribute, or try the tool yourself to see how wind energy can work for you.

Link : https://windcalc-kkvnhe3y.manus.space