I'm a first-year BTech CS student at TSDCEM in Vasai. Before college started, I was thinking about what problems actually exist on campus — not hypothetical startup problems, but things that would genuinely annoy someone in their first week of college.

The answer was obvious once I thought about it: students constantly need small things done that they either can't do themselves or don't have time for. Edit a reel for a college fest. Explain a concept before an exam. Design a poster. Fix a bug in an assignment. These aren't big tasks. They don't need a freelance platform with contracts and portfolios. They just need a quick connection to someone nearby who can help.

The existing options were wrong for this

I looked at what existed. Fiverr is too formal — it's built for professional work, not for "help me understand recursion before tomorrow's exam for ₹100." WhatsApp groups sort of work but have no trust layer, no payment structure, and no way to find someone with the right skills. Word of mouth works until it doesn't.

What I wanted was something that felt more like asking a friend — hence the name. You post what you need, someone who can do it accepts, you pay a fair price, and both of you walk away with something useful.

What I actually built

The core loop is: post a task with a price (minimum ₹50), a verified helper accepts it, they deliver the work, you approve and pay the second half via UPI.

The trust system was the part I spent the most time on. There are two separate scores — a Helper Score that only goes up (your track record) and a Reliability Score that starts at 100 and drops if you cancel, miss deadlines, or get disputes upheld against you. Below certain thresholds, you can't take higher-paying tasks until you rebuild your score. This mirrors how Free Fire's honour system works — and it turns out that's a mechanic college students already understand intuitively.

Skill verification was another deliberate choice. Before a helper can take tasks above ₹200 in a category, they submit a short sample (a 10-second clip for editing, a writing sample for proofreading). It filters out people who claim skills they don't have.

The book trade center

One thing I added that wasn't in the original plan: a campus book exchange. Students pay full price for textbooks they use for one semester and never touch again. So I added Buy, Trade, and Lend listings for books — with condition standards, negotiation support, and the same UPI payment flow as tasks. It felt like a natural extension of the "campus economy" idea.

What I learned

The hardest part wasn't the code — it was designing systems that couldn't be gamed. A marketplace where trust is the product means every feature has to be thought through from the perspective of someone trying to abuse it. That thinking changed how I approach every design decision.

The site is live at help.thefixfy.com. I'm running a closed beta with students at TSDCEM before opening it up.

I built FinMG and Help A Friend in parallel — which means I made the classic mistake of designing them independently and ending up with two apps that looked like they came from completely different companies.

FinMG had a dark glassmorphism aesthetic with purple accents. Help A Friend was monochromatic black and white. Both had different fonts, different spacing, different button styles. When I looked at them side by side, they felt unrelated.

Why it mattered

If Thefixfy is going to be a real company with multiple products, those products need to feel like they belong together. Users who find one product through the other should feel a sense of familiarity — the same way Google's products feel cohesive even when they serve completely different purposes.

A shared design system also means less work going forward. Every time I build something new, I shouldn't be making basic decisions about font sizes or border radii from scratch.

The decisions that mattered most

Typography first. I settled on Space Grotesk for headings and Inter for body text across both products and the company site. Space Grotesk is geometric and confident — it reads as modern without being trendy. Inter is the most legible screen font at body sizes. This decision alone brought the two products significantly closer together.

One color palette, adapted per product. The base palette is strictly monochromatic: #FAFAF9 (off-white), #111110 (near-black), and a warm grey family in between. Both themes (light and dark) are defined as CSS variables, so every component inherits the right colors automatically.

FinMG adds green for positive financial signals and red for negative ones — these are meaningful colors in a finance context, not just decoration. Help A Friend uses the monochrome palette only, since there's no financial data that needs color coding.

Spacing as a system. I standardized card padding at 24px, section vertical gaps at 48–96px, and border radius at 12px for cards and 9999px (full pill) for CTAs. These numbers sound arbitrary but they're not — they create a consistent visual rhythm that makes pages feel considered rather than assembled.

What I'd do differently

Start with the design system before building the first component. I wasted time retrofitting CSS variables into code that had hardcoded hex values everywhere. If you're building more than one product, write the CSS variable definitions on day one and reference them everywhere from the start.

The system is still evolving — but both products now feel like they came from the same place. That's the goal.

Both Help A Friend and FinMG run entirely on Supabase. Auth, database, realtime subscriptions, file storage — all of it. Here's an honest take on that decision.

What Supabase gives you

The obvious thing is that it's a Postgres database with a REST API auto-generated from your schema. But the parts that actually matter day-to-day are the ones that would take weeks to build yourself.

Row Level Security. This is the biggest one. RLS lets you define access rules at the database level — "a user can only read their own financial data," "only the task poster or helper can read messages for that task." These rules enforce themselves regardless of how a request comes in. You can't accidentally write a query that returns another user's data if RLS is on.

Auth that actually works. Email/password, Google OAuth, and session management — all handled. The JS client refreshes tokens automatically. I didn't write a single line of auth infrastructure.

Realtime out of the box. Help A Friend's messaging system uses Supabase Realtime — new messages appear instantly without polling. Same for new task listings appearing live in the browse page. One channel subscription, a few lines of code.

Storage with the same RLS logic. Task deliverables (private, only poster and helper can access), skill verification samples (reviewed by admins), and portfolio items (public) all live in Supabase Storage with policies that mirror the database access rules.

Where it gets complicated

The free tier limits are real. 500MB database, 1GB storage, 50MB file uploads. For a beta with a small number of users, this is fine. For anything serious, you're looking at $25/month minimum. Plan for this before you launch to real traffic.

Edge Functions have cold starts. I use a Supabase Edge Function to proxy Gemini API calls for FinMG's AI features (keeping the API key server-side rather than exposing it in the client bundle). Cold starts on the free tier can add 600–800ms to the first AI request. Not a dealbreaker, but something to know.

Migrations need discipline. Because Lovable generates code and I'm also writing SQL directly in the Supabase dashboard, keeping migrations in sync requires paying attention. I've had to re-run a migration manually twice because the generated code assumed a column that hadn't been added yet.

Would I choose it again

Yes, for a product at this stage. The speed advantage is significant — I shipped the entire Help A Friend backend (auth, tasks, scoring, messaging, payments, file handling, disputes) without setting up a single server. For a student founder building multiple products simultaneously, that matters a lot.