WealthPilot

Overview

WealthPilot reads a Zerodha account through Kite Connect and answers the question the broker's app can't: will each goal make it? The analytics engine is plain Python functions with no framework in them, so it's tested on its own. The one CPU-heavy job, simulating thousands of market paths per goal, runs in a C++20 service behind gRPC instead of in the request path. Around it: a daily snapshot job, a live price stream, and an access token that expires every day and is detected and recovered instead of failing. The public demo runs on a fictional investor, read-only.

More screens

Problem → built → result

Problem

A broker app shows balances. It can't say whether a goal will actually be met.

Built

Next.js on async FastAPI, a C++20 Monte Carlo engine over gRPC, a rules engine for risk.

Result

180 backend tests · 99% engine coverage · live demo

Features

• Monte Carlo goal simulation in a multi-threaded C++20 service behind gRPC, with an in-process fallback so the backend runs and tests without it
• A pure analytics engine (bucket classification, risk rules and goal glide-paths) at 99% line coverage, flagging concentration, sector clusters and drawdowns
• Kite Connect's daily token expiry is detected, surfaced and recovered; a scheduled job snapshots the portfolio each weekday
• Orders only through a human-in-the-loop sell flow: a dry run by default, placed only on explicit confirmation
• A public demo mode that serves a fictional portfolio read-only: every write is refused and every key dropped, even if the environment sets one

Decisions

1. 1

   Move only what's CPU-bound.

   Simulating thousands of market paths per goal is the one heavy job, so only it runs in a multi-threaded C++20 service behind gRPC. An in-process fallback keeps the backend running and testable without it.

2. 2

   Keep the engine free of frameworks.

   Bucket classification, risk rules and goal glide-paths are plain Python functions with nothing to mock, which is what keeps 99% line coverage cheap to hold.

3. 3

   Expect the token to expire.

   Kite Connect's access token expires every day, so expiry is a normal state that's detected, surfaced and recovered, not an error the page falls over on.

4. 4

   A human places every order.

   The sell flow is a dry run by default and places an order only on explicit confirmation. The public demo goes further: every write is refused and every key dropped.