A founder's story · Mumbai, India · 2026
How the loneliest months of a job hunt became a mental wellness app for India's most stressed generation
Origin
It started in early 2026. The job hunt had been going on long enough to get heavy. Rejections, silence, more rejections. The anxiety had started bleeding into everything: sleep, relationships, how I felt about myself at 11 PM on a Tuesday. I was sitting alone with no one to call and no safe place to put the thoughts that were piling up.
I tried ChatGPT. It answered my questions. It was technically helpful. But it wasn't with me. Every exchange felt like a search result. No warmth, no thread from what I'd said five minutes ago, no sense that anything was listening rather than responding.
I kept thinking about everyone I knew going through versions of the same thing. The college student before placements. The young professional dreading their first appraisal. The job seeker who's stopped telling their family how bad it is. India has 600 million people under 25. Mental health infrastructure is thin, expensive, and still heavily stigmatised. The gap was hard to ignore.
On March 9, 2026, I opened a blank file and started building.
The Thesis
Indian youth aren't avoiding mental health support because they don't need it. They're avoiding it because the options feel clinical, expensive, and designed for someone else. A product that earns trust first, before asking for anything, could actually work.
Target User
Urban Indian, 18–35. Working professional or college student. Fluent in English and Hinglish. Under pressure from career, family, and the constant social-media performance of looking fine.
The Name
Aruveli is a coined name, built from two languages. Aruna (Sanskrit) is the soft reddish glow of dawn. Veli (Tamil) is companion, friend. Together: the dawn's companion. The light after the dark. The tagline: "here before the light."
The Builder
No technical co-founder. No design team. No product budget. I did everything myself from day one. Honestly, that turned out to be the most valuable part of the whole process. You learn what each role actually involves when there's no one else to pass it to.
Product Manager
Defined the product vision, roadmap, and feature prioritisation. Ran user research via a structured 36-response survey. Made hard calls about what not to build. Translated raw user pain into product specs.
UX / UI Designer
Designed all user flows, interaction patterns, and the visual identity. This covers the Aruveli flame mark, the typography system, and the dark terracotta colour palette. The goal was a product that feels intimate, not clinical.
Frontend Developer
Built the entire product in HTML, CSS, and JavaScript with no framework. Integrated Supabase for auth and data persistence, deployed on Cloudflare Pages, wired the Anthropic API directly, and debugged every broken interaction myself.
AI / Prompt Engineer
Designed Sofia's conversational personality: 6 distinct emotional modes, crisis detection logic, and the system prompt architecture that keeps her feeling like a companion. The work was mostly about language, not code.
DevOps
Set up and maintained the full deployment pipeline. Cloudflare Pages for hosting, Supabase for the backend, GitHub for version control, and environment configuration across dev and production. Managed deploys, debugging, and uptime solo.
The Build
No prior coding experience. I built Aruveli from a blank file. Every feature was researched, designed, written, debugged, and shipped by one person. Here's the honest sequence of how it came together.
March 9, 2026
Day 1: The First Prompt
Started with a simple idea: an AI companion that listens differently from ChatGPT. The first prototype was rough, a chat window with a name. But the intent was clear enough to keep going.
Week 1
Product Definition & Research
Drafted and distributed a 12-question user survey. 36 responses came back. Career pressure dominated stress at 75%. Therapy was out of reach for most respondents, either too expensive or too stigmatised. The single most requested feature was an AI chat companion available at any hour.
Week 2
Core Features + Identity
Built Sofia AI chat with 6 emotional modes, the plant XP system, mood chips, breathing exercises, an anonymous community feed, crisis detection, and voice mode. Also designed the brand in parallel: the Aruveli flame mark, terracotta palette, Cormorant Garamond typography, and the tagline "here before the light" (the dawn's companion, the light after the dark).
March 24, 2026
Live on aruveli.com
Deployed index.html and app.html on Cloudflare Pages. Supabase backend live with chat persistence. Beta allowlist auth active. PWA-ready with favicon and web manifest. The product is live and usable today.
The Product
Aruveli isn't a therapy app or a meditation app. It's a listening companion: something you open when you're not okay but not ready to call anyone. Here's what's built:
Sofia AI Chat
An AI companion powered by Claude Sonnet. Responds in 6 emotional modes matched to how you're feeling. Not Q&A. Actual conversation.
Plant XP System
Your emotional consistency grows a virtual plant through 7 levels. Gamification as self-care reinforcement. Check in, it grows.
Mood Chips
Express how you're feeling in one tap. Radiant, Good, Calm, Tired, Anxious, Low. Sofia adapts her entire tone to match.
Breathing Exercises
Box breathing, 4-7-8, and Natural rhythm. Guided, timed, and interruptible. Built for 3 AM anxiety spirals.
Anonymous Community
Share feelings without your identity. Auto-assigned codenames keep it safe. See others' posts. Know you're not alone.
Voice Mode
Talk to Sofia out loud using Web Speech API. She speaks back. For when typing feels like too much effort.
Crisis Detection
Detects crisis-level language in chat. Instantly surfaces iCall and Vandrevala Foundation helplines. Responsibility built in, not bolted on.
Weekly Letter
AI-written reflection on your week. Patterns Sofia noticed, gentle observations, something to look forward to.
8 Badges + Streaks
Recognition for showing up. Not for being fine, but for trying. Streaks, milestones, quiet acknowledgements of effort.
Sofia doesn't have one personality. She has six. Each maps to a mood state and changes her tone, language, and response approach to match.
Radiant
Energetic, celebratory, enthusiastic. Meets you in your best moments without diminishing them.
Good
Warm, steady, present. The everyday companion mode. Comfortable and grounded.
Calm
Gentle, slow, unhurried. For moments of quiet or winding down. No urgency.
Tired
Soft, compassionate, low-demand. Doesn't push. Just holds space when you're running on empty.
Anxious
Reassuring and grounding. Acknowledges the spiral without feeding it. Breathing cues, gentle redirects.
Low
The most important mode. Non-judgmental, unhurried, present. No toxic positivity. Monitors for crisis signals.
The Research
Before writing any final code, I ran a structured user research survey. The data shaped every feature decision, the pricing model, the language, and the tone. Here's what came back.
36
Survey Responses
75%
Cited Career / Job Pressure as #1 Stressor
~70%
Feel Stressed Daily or Multiple Times a Week
100%
Who Answered the Beta Question Said Yes
Biggest Sources of Stress (multi-select)
What Features Matter Most (pick up to 3)
Why Haven't They Tried Therapy?
Most people haven't ruled out getting help. They're just not convinced anything would actually work. Aruveli's low-barrier entry is designed to close that trust gap first.
Key Demographics
94.4% based in metro cities (Mumbai, Delhi, Bangalore)
77.8% are working professionals
52.8% aged 23–27 (core target bracket)
61% prefer English, 36% prefer Hinglish
Only 8.3% currently see a therapist
Willingness to Pay: The Honest Finding
47% said they'd only use it if it's free. That's not a failure. That's the data telling me how to sequence things. The product earns trust first through a generous free tier, then converts through real demonstrated value. Every consumer wellness app in India that's worked has followed some version of this.
The Hard Parts
Building Aruveli as a non-technical founder meant every hard problem was a chance to stop. Here are the ones that pushed hardest.
Making an AI feel like a companion, not a search engine
The hardest part wasn't the code. It was the language. Getting Sofia to respond in a way that felt like being heard took dozens of prompt iterations. Too formal and she felt robotic. Too casual and she felt hollow. Finding the right voice took weeks of testing different system prompt structures, tone guides, and response frameworks.
Feature scope: knowing what to cut
I built and scrapped more features than made the final cut. Every idea felt important when I had it. The discipline of saying "not yet" on things like therapist matchmaking, journaling AI analysis, and detailed mood analytics came from forcing myself back to one question: does this help someone feel less alone right now?
Building without knowing how to code
Every feature was built through iteration: describe what I wanted, understand the output, break it, fix it, figure out why. That process turned me from a product thinker into someone who actually understands the stack. Supabase schema design, Cloudflare Pages deployment, API key management, localStorage architecture. I own all of it now.
Handling crisis responsibly
When I decided to build crisis detection into Sofia, I had to think seriously about what responsible AI in mental health actually looks like. Aruveli doesn't pretend to be therapy. The crisis detection layer surfaces human support immediately (iCall and Vandrevala Foundation) and never pulls the AI deeper into the conversation. Getting that boundary right mattered more than any other feature decision.
Finding a name that felt alive
The naming process took longer than expected. I tested ZenFlow, Unfurl, Exhale, Unsaid, 3AM. Each captured something but not everything. Aruveli came from a different direction entirely: a coined word built from two languages. Aruna (Sanskrit) for the soft reddish glow of dawn, and Veli (Tamil) for companion or friend. Together: the dawn's companion. It stuck because it was the only name that felt like the product's identity rather than a label attached to it.
What I Learned
Trust before monetisation. Always.
47% of respondents said they'd only use the app if it's free. That's not a user problem. It's a sequencing problem. The product has to prove itself before asking for anything. Pricing follows trust, not the other way around.
Research shapes product better than intuition
My instinct was to make journalling a centrepiece. The survey showed only 27.8% actually wanted it, while anonymous community and breathing exercises both scored 50%. Without the data, I would have built the wrong thing very confidently.
AI as a thinking partner, not a shortcut
AI tools didn't write Aruveli for me. Having one available at 2 AM to think through problems helped move faster than I could have alone. But the decisions, the testing, the debugging, the product judgment: all of that was mine. The tools compressed the learning curve but didn't replace the thinking.
The emotional problem is harder than the technical one
Getting the auth flow to work took a day. Getting Sofia to say the right thing when someone is having a hard night took weeks. The hardest product problems aren't engineering problems. They're empathy problems. Tone problems. The courage to not over-engineer things that should just feel human.
Survey demographics are not your user demographics
My early survey reached mostly founders and working professionals: a higher-stress, higher-resilience group than the average target user. The responses were still useful, but I learned to read them with that in mind. Real users will be more emotionally raw. The bar for trust is higher than the data suggests.
The Why
~200M
Indians estimated to have a mental health condition
0.75
Psychiatrists per 100,000 people in India. WHO norm is 3 per 100,000.
₹1,500–3,500
Typical cost of a single private therapy session in metro cities
Mental health in India is a supply problem, a stigma problem, and a cost problem all at once. Therapy is expensive and hard to access. Most young people dealing with something serious have never spoken to a professional. Not because they don't want to, but because the whole thing feels either impossible or unnecessary.
The Aruveli Thesis
The product doesn't need to save lives to matter. It needs to make the quiet hours of someone's night feel a little less heavy.
The go-to-market isn't advertising. It's proximity. College campuses during placement season. WhatsApp communities. LinkedIn posts from founders who've been through a hard stretch. Referrals from the 36+ people who've already said they want in. The numbers to hit breakeven are small. The kind a founder who's already this close can actually reach.
The community product itself is the distribution mechanism. Anonymous stories shared on Aruveli's feed become the social proof that makes someone else download it at 3 AM.
A live product, a researched market, a founder who built it from their own need and kept going.
Aruveli is live at aruveli.com. Sofia is ready to talk. If you've felt the 3 AM loneliness this product was built for, it was built for you.
linkedin.com/in/mihir-kothari-718927259
Aruveli · here before the light · © 2026 Mihir Kothari </