The problem
A stock advisory product has two hard constraints that fight each other: the data has to be live, and the advice has to be auditable. Traders act on a number that was true one second ago; regulators ask what the number was last Tuesday.
There was no product when I joined. There was a thesis, three engineers including me, and a market data feed.
The approach
Two paths through the same data. A hot path — Socket.io over Redis pub/sub — pushed live prices to whoever was watching. A cold path wrote every tick and every advisory action to durable storage, so any recommendation could be reconstructed with the exact prices that produced it.
Everything expensive moved off the request path into BullMQ and AWS SQS. That is what let a small team absorb growth without a rewrite: the user-facing surface stayed thin while the work behind it queued up and drained.
What shipped
- The core platform, from zero to 100,000+ users
- A Chrome extension that passed 10,000 installs, bringing the advisory feed into the sites traders already used
- Real-time price processing on Socket.io and Redis
- Automated test pipelines that cut bug reports by around 70%
I led three developers across frontend and backend for most of that run.
What I would do differently
When SEBI tightened its rules on advisory platforms, the funding wobbled and the runway shortened. I kept working for two months without pay because the users were still there and the system still needed someone.
I do not regret it, but I learned to separate two questions I had been treating as one: is this worth building and is this worth building for free. They have different answers, and you should answer them separately, in advance.