Elytra

Elytra — Stop Guessing & Start Building

Inspiration

It started with frustration — the kind every founder knows too well. You’re trying to grow your startup, but end up drowning in tabs comparing Zendesk vs. Intercom vs. Help Scout. Reviews are biased. “AI” chatbots spit out generic suggestions. You burn weekends testing tools that don’t fit.

We realized something simple: everyone gives recommendations, but no one gives proof.

At CalHacks, we decided to fix that. Our goal? Build an AI that doesn’t tell you what to buy — it shows you why.

Turn software choices from guesswork into math. Turn software choices from guesswork into math.

That was the spark that became Elytra — the first AI tool recommender that connects to your actual data, simulates how a tool performs, and proves ROI in seconds.

What it does

Elytra helps startups make smarter, faster, and data-backed software decisions.

You tell it what’s slowing you down — your problem area, team size, budget, and the tools you already use. Elytra’s AI then analyzes your setup and produces a personalized recommendation report with:

Top 3 tools ranked by fit score (out of 100)

Estimated time saved and efficiency boost
ROI payback period calculated from your inputs
Compatibility checks — API access, integrations, setup risks
Implementation guide with rollout steps and best practices

And here’s the part everyone loves:

You can test a tool instantly. Elytra connects to your Slack or Gmail, runs a live simulation, and shows the workflow in action — all with your real data, in under 8 seconds. No hypotheticals. No fluff. Just proof.

How we built it

We built Elytra as a full-stack TypeScript app with a focus on speed, clarity, and reliability.

Frontend: Next.js 14 + Tailwind + shadcn/ui — polished, responsive, and fast

Backend: Next.js API routes with Prisma ORM + SQLite

AI: Anthropic Claude 3.5 Sonnet — powering ROI predictions and reasoning

Integrations: Slack OAuth for live simulations, Puppeteer for dynamic PDF reports

Deployment: Vercel with automated CI/CD for instant updates

Our workflow was classic hackathon madness — caffeine, chaos, and collaboration. We split responsibilities: one on integrations, one on the frontend, one on backend logic. We lived in VS Code and GitHub branches, merging code at 3 a.m. and high-fiving over bug fixes.

48 hours later, Elytra was live. Not a prototype — a fully working system that could recommend, simulate, and prove results.

Challenges we ran into

Frontend–Backend integration:Connecting APIs across Vercel and Render gave us serious CORS nightmares. Solved with persistence (and too much coffee).

OAuth setup: Slack’s OAuth flow took multiple retries to get right, but when it finally posted a real ticket message in #support — that was our “wow” moment.

LLM orchestration: Getting structured, reliable outputs from an AI model for ROI math took trial, error, and a few philosophical crises.

PDF generation: Puppeteer + Vercel wasn’t playing nice, so we wrangled Chromium binaries manually. Worth it for the clean, sharable reports.

Accomplishments that we’re proud of

Shipped a complete full-stack product in under 48 hours
Integrated live Slack OAuth and posted real messages in real time
Built a data-driven ROI engine with Anthropic’s Claude
Designed a polished UI that feels modern, not like a chatbot
Learned an entirely new stack — and shipped anyway
We didn’t just build a hackathon demo — we built something people can actually use.

What we learned

We learned how to work as a team under pressure — sharing code, debugging in sync, and pushing through exhaustion.

Technically, we mastered:

GitHub collaboration and version control
Linking TypeScript frontends to AI-powered APIs
Structuring LLM prompts for reliable, parseable output
Turning abstract AI reasoning into real, measurable results
But beyond the code, we learned how to build something that feels human — software that proves value instead of promising it.