StackAdvisor — Build Log

Started with research before touching any code

Brief says research must come before code. Most candidates will skip this. I spent the first 2 hours finding 3 real Indian startups with public engineering documentation. Chose BrowserStack, Niramai, DeHaat — all uncommon picks that most people won't cite (everyone else will use Razorpay + Zepto).

Avoided Razorpay, Zerodha, Zepto — those are the first names every candidate will reach for. Differentiation starts here.

Niramai card originally said AWS — had to correct it

My first draft of the Niramai card listed "AWS GPU instances" for hosting. This was wrong. The actual source — cloud.google.com/customers/niramai — is a Google-authored case study that explicitly confirms GCP + GKE + TensorFlow.

Important: a judge can verify this in 10 seconds. Getting it wrong would have been an immediate credibility hit during the live debrief. Lesson — always cite the source first, build the claim second.

DeHaat stack had two wrong claims — fixed via LinkedIn

Initial card said Java Spring Boot + MySQL for DeHaat backend and database. Found the actual stack on the LinkedIn profile of a current DeHaat SDE2 (Core Backend Engineering, Supply Chain domain, Jan 2025–Present): Python/Django + Go + PostgreSQL + Amazon S3 + Redshift + RabbitMQ + Celery + Kubernetes + Elasticsearch + Kibana.

A current employee's own profile is the strongest possible source because it reflects what they actually built in production, not what was listed in a hiring ad.

Decided to build chat-style UI instead of a dropdown form

Every other candidate will build a 4-dropdown form with a result card. That's what AI tools default to. Instead, I'm building a chat-style interview — the tool asks one question at a time, like a real advisor would. Same 4 inputs (stage, team, budget, sector), but the interaction pattern is completely different.

Trade-off: chat UI is harder to build in vanilla HTML/CSS/JS. But the brief says no frameworks — which means everyone's tool will look similar if they don't take a structural risk. Taking the risk here.

Changed color palette from neon green to dark indigo + mango amber

First version used #050505 background + #00ff66 neon green — the single most common "terminal aesthetic" combination. It's what AI tools default to. It reads as AI-generated immediately.

Reason: mango amber is distinctive, doesn't feel robotic, has warmth, and zero other candidates will have it. Also subtly references India's context without being literal about it — which fits a tool built for Indian startups.

Reset button was doing a full page reload — killed the conversation

First version of the reset button used onclick="location.reload()". This cleared everything — the entire chat history, all user answers, both panes. Not the right UX. A founder who wants to try a different combination shouldn't lose their conversation.

Small thing but the kind of detail that matters in a live debrief where the judge might test it.

Built the 10x Scale Toggle with cost delta — not just flash-red

The Scale toggle idea was originally "flash components red when they break." I upgraded it: each breaking component now shows the exact cost delta — not just a warning colour.

This is the most powerful moment in the tool. It quantifies what "scale" actually costs in INR. Makes systems thinking visible to a non-technical founder who might not know what "hitting the free tier limit" means in real money.

Also added a recalculated total cost that updates when the toggle fires — so the ₹X/mo figure at the top changes, not just the component cards.

Added Graveyard card — the most original element in the tool

For every combo, added a real Indian startup that picked the wrong stack at that stage and what it cost them. Real failures, not hypotheticals.

Founders respond to loss aversion more than opportunity. Seeing what went wrong for someone else in their exact situation is more persuasive than any best-practice recommendation.

Verified all INR cost numbers against live pricing pages

Went through every cost figure in all 6 combos and cross-checked against the actual pricing page, not estimates. Used $1 = ₹84 conversion throughout.

Every number in the tool now links to the actual pricing page URL. Clicking any source takes you directly to the line item that proves the number.

Fallback logic was broken — always returned EdTech combo

The fallback for unmatched combos was recommendations[Object.keys(recommendations)[0]] — which always returned the first key in the object regardless of what the user selected. A founder who picked HealthTech + Scale would get an EdTech stack. That's a serious UX failure.

Also added a visual note in the result card when a fuzzy match is used, so it's transparent that this isn't an exact match.

Pre-submission checklist run — 3 gaps found and fixed

Did a final pass through the brief requirements against the actual tool. Three gaps found:

All 3 Indian startups documented with sources that a judge can open and verify in under 60 seconds: BrowserStack (Vault blog), Niramai (GCP case study), DeHaat (YourStory CTO interview + LinkedIn SDE2 profile). Contrarian picks flagged with startup evidence. INR costs linked to live pricing pages. 6 combos with genuinely differentiated outputs.