PortfolioPulse: Real-Time Multi-Asset Portfolio and Alert Engine

Holdings are scattered across brokers (Groww, Zerodha, Upstox) with no single view of net worth or true profit and loss

In many Indian households one person tracks the whole family's investments, so a spouse's or parents' holdings end up mixed into the same spreadsheet with no clean way to keep them apart

Stock-only trackers ignore half the portfolio. Indian investors also hold ETFs, mutual funds, and physical gold and silver, invisible in most tools

Broker alerts are a single ping, not a journey. There is no staged notification as a position moves from entry toward its target, or drifts toward its stop

Watching dozens of positions by hand is error-prone. The one moment that matters, the touch of a target or a stop, is the one you miss

Alert noise trains you to ignore alerts. Pings fire for watchlist names you do not even own

Indian mobile networks drop. A cloud-only app becomes unusable on a flaky connection

Fire-exactly-once staged alerts under concurrent evaluators

Each position moves through buy-range, halfway-to-target, break-even, approaching-stop, target, and stop. The terminal target and stop pushes fire exactly once, even when the intraday poller and the end-of-day evaluator both decide the level has been reached at the same instant, guarded by an atomic compare-and-set on the alert row. The progress milestones in between are de-duplicated through a trigger-history ledger, so each stage lands a single time.

Race-safe holding recompute on the money path

Every holding recompute runs inside the write transaction under a PostgreSQL advisory lock, so concurrent edits to the same portfolio serialize instead of interleaving. Sells are checked against units held by a race-safe oversell guard, and transaction sync is idempotent through a client-supplied id backed by a unique-indexed upsert, so a retried request cannot double-apply.

Configurable two-stage trailing stop

Protection ratchets up in two stages as a trade works in the holder's favour and never loosens. The default was stress-tested across roughly 23 years of NSE price history (about 98,000 simulated trades) and outperformed the flat and volatility-scaled alternatives tested. This was an engineering-rigor exercise to validate a design default, not a promise of returns.

Offline-first sync over flaky networks

A local Drift and SQLite cache keeps the app fast and readable on a poor connection. It reconciles with the server through a debounced sync engine that runs periodically, on reconnect, and after edits, and the intraday evaluator rejects stale or outlier ticks before they can trigger a level.

Realtime pricing with a hard fallback

Realtime quotes come from Zerodha Kite Connect and stream to the app over SignalR, so the consolidated portfolio recomputes live P&L with no user action. Yahoo Finance is the fallback source, and end-of-day ingestion retries and fails loudly instead of silently writing stale data.