Projects — The Portfolio Ladder
Projects are what land interviews. The depth of understanding behind them is what lands offers. And here is the uncomfortable truth about most portfolios: they are noise. Todo apps. Weather dashboards. URL shorteners. The tutorial-followed “Netflix clone” with someone else’s architecture and someone else’s bugs. Recruiters at Razorpay and PhonePe have seen ten thousand of each. They signal exactly one thing — this person can follow a tutorial — and that is a skill a prompt mass-produces for free.
This section is the antidote: four projects, in order, each one a problem real companies pay real engineers to solve. You have finished all eight tracks on this site. These projects are the exam.
What Makes a Project Signal
A portfolio project is worth building only if it passes all four tests:
| Test | What it means | What fails it |
|---|---|---|
| It solves a paid problem | You can name the job title and team that owns this problem at a real company | ”It’s like Twitter but for X” |
| You can defend every decision | Why this matching key, why this table shape, why a map and not a sort — answered without notes | Generated code you read like a stranger |
| It is deployed and demoable | A stranger can see it run in two minutes — live URL or a demo script that just works | ”It works on my machine, let me fix one thing” |
| It generates true numbers | Resume bullets with metrics you measured: rows processed, milliseconds taken, failures recovered | ”Improved performance significantly” |
If a project idea cannot pass the first test, do not build it. That single filter kills the todo app, the clone, and the URL shortener instantly — nobody is paid to build those.
The Ladder
Five rungs. The first one you have already climbed.
| # | Project | The real problem | Duration |
|---|---|---|---|
| 0 | SplitEase | Group expenses, paise-exact math, settle-up — built across the tracks as the CLI capstone and the REST API capstone | Done ✅ |
| 1 | Recon — the reconciliation engine | The bank file says X, your ledger says Y, and the gap is lost money someone must explain by 9am every morning | ~2 weeks |
| 2 | Relay — a persistent task queue | Every async system — emails, exports, webhooks — needs jobs that survive a crash, retry with backoff, and never run twice by accident | ~2–3 weeks |
| 3 | Dispatch — a notification service | One event, many channels, per-tenant rate limits — and a duplicate notification is a support ticket, so one event must never send twice | ~2–3 weeks |
| 4 | RupeeRail — a mini payment switch | Payment intents, flaky bank adapters, a double-entry ledger where balances always sum to zero — the flagship that integrates everything | ~4–6 weeks |
SplitEase already proved you can build a secured, validated, tested API. The ladder proves you can build infrastructure — the systems product features stand on.
Why This Ladder
Three deliberate properties, none accidental:
Each project consumes the previous one’s patterns. Recon’s idempotent-batch discipline reappears in Relay’s job claims. Relay literally becomes Dispatch’s engine. RupeeRail uses Dispatch’s webhook patterns for merchant callbacks and Recon’s matching for its settlement report. By project 4 you are not learning patterns — you are reusing your own, which is what senior engineers actually do.
Together they cover the whole site. File processing and streams, hash-map matching, SQL transactions and locking, concurrency, retries and timeouts, rate limiting, idempotency, double-entry money. Every track you finished gets cashed in. Each project doc maps its concepts back to the exact modules where you learned them.
They are all fintech-adjacent — on purpose. Your target market is backend roles at Indian product companies, and the densest cluster of those roles is payments: Razorpay, PhonePe, CRED, Juspay, and everyone who integrates with them. Why fintech is the target makes the full case. A portfolio that says “reconciliation, task queues, notifications, payment switch” reads like the org chart of a payments company’s platform team — because it is.
The Rules
These are not suggestions. Breaking them is how the ladder turns back into portfolio noise.
- One project at a time. Two half-finished projects signal less than zero. Finish, deploy, document, then start the next.
- Design on paper first. Every project doc has a Design It Yourself First section — numbered questions you answer in writing before reading the reference design. This is mandatory. Design judgment is the skill that survives the AI era; the reference design exists to check your answers, not to replace them.
- Deploy everything. A project that only runs on your laptop is a claim, not evidence. Free tiers only — the same Railway/Render/Fly.io options from week 9.
- Measure before you write resume bullets. Every project ends with bullets containing placeholders — N records, M seconds. You fill them in from real runs on your real code. A number you did not measure is a lie you will be caught telling in an interview.
- Every project gets a README that could onboard a stranger. What it is, why it exists, how to run it, the key design decisions and why. If a stranger cannot go from
git cloneto a working demo using only the README, the project is not done.
How Projects Fit the Schedule
The ladder starts after the 12-week sprint, or alongside month 3 if the tracks are done and you want overlap. Either way, the honest math at your real pace of 2–4 focused hours a day:
| Project | Sittings of 2–4 hrs | Calendar time |
|---|---|---|
| Recon | ~10–12 | ~2 weeks |
| Relay | ~14–18 | ~2–3 weeks |
| Dispatch | ~14–18 | ~2–3 weeks |
| RupeeRail | ~25–35 | ~4–6 weeks |
That is roughly three to four months for the full ladder — and that is fine. You do not need all four to start interviewing. Recon alone is a stronger portfolio than most candidates bring, and each additional rung compounds it. Interview with what you have; keep climbing while you do.
The trap to avoid: rushing RupeeRail to “have the flagship” while Recon sits half-deployed. A finished rung beats a started ladder. Every time.
Start Climbing
Project 1 is waiting: Recon — the reconciliation engine. It is two weeks of work, it produces your first measured resume number, and it solves the problem every payments company in India wakes up to at 7am.
Go design it.