Drift

A smarter watchlist for what actually changed.

Drift turns market movement into clear, explainable attention signals so you can spend less time scanning prices and more time understanding what deserves a closer look.

Try the live demo · Watch the walkthrough · See the technical explanation

A normal watchlist tells you that a price moved. It does not tell you whether that move is unusual for the stock, whether it happened with meaningful volume, whether it is close to a 52-week extreme, or whether you have already reviewed it.

Drift is built around that missing layer of context. It remembers where you last looked, compares new market data against each symbol’s own behavior, and surfaces the changes that are most worth investigating.

Drift is not trying to predict the market. It helps you decide what deserves your attention next.

Drift separates meaningful changes from normal movement. The overview is organized into clear attention tiers, so the most relevant symbols rise to the top instead of getting buried in a long ticker list.

Every surfaced item has visible evidence. Drift can flag abnormal movement, unusual volume, 52-week context, portfolio-wide movement, and behavior that differs from the benchmark. The rules are readable, inspectable, and designed to be audited.

The key comparison is not only “what happened today.” Drift remembers the price and time associated with your last real review, then shows what changed since you were actually there.

Create a watchlist from scratch, add symbols to one or more lists, or start with a premade collection for themes such as technology, banking, pharma, EVs, and more.

Explore historical ranges with hover inspection, price changes, OHLC data, volume, range returns, and candlestick or line views where the underlying data supports them.

Review what Drift has surfaced over time, mark items as seen, and save a short thesis for why a symbol is on your list. The product keeps both the signal and your own reasoning in view.

Drifty AI can turn already-computed signals into a concise explanation. It does not decide what gets flagged; the deterministic rule engine does that first.

A symbol can enter the attention feed when one or more transparent checks fire:

1. Abnormal movement: the move is large compared with the symbol’s recent behavior.
2. Unusual volume: trading volume is meaningfully above its recent average.
3. 52-week context: price is approaching or crossing a relevant yearly high or low.
4. Portfolio context: several symbols in the watchlist are moving together.
5. Market context: the symbol is behaving differently from its benchmark.

These rules contribute to an attention score and produce a readable reason. Drift does not present an opaque recommendation or pretend certainty about the future.

1. Open the live demo.
2. Complete the two short onboarding questions.
3. Choose a premade watchlist or create a new one.
4. Search for and add symbols such as NFLX, DIS, or UBER.
5. Return to Overview to see market context, tracked symbols, quiet movement, and attention items.
6. Open a symbol to inspect its reasons, chart, thesis, and available actions.
7. Mark it as seen to establish a personal review baseline.
8. Use Charts, History, and the optional Drifty AI explanation when you need more context.

Drift has a React and Vite frontend backed by a FastAPI service.

• Python 3.11 or newer
• Node.js 20 or newer
• npm

cd backend
python -m venv .venv

# macOS / Linux
source .venv/bin/activate

# Windows PowerShell
.venv\Scripts\Activate.ps1

pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

The backend runs at http://127.0.0.1:8000. SQLite is created automatically on first run. Set DATABASE_URL if you want to use PostgreSQL instead.

In a second terminal:

cd frontend
cp .env.example .env
npm install
npm run dev

Open http://localhost:5173.

The frontend uses VITE_API_BASE=http://127.0.0.1:8000 by default. Adjust the local environment file if your backend runs elsewhere.

The product works without an AI key. To enable optional plain-English summaries, add one or more comma-separated GROQ_API_KEYS values to backend/.env. AI only rephrases signals that Drift’s deterministic engine has already computed.

React + Vite
     │
     │ polls for watchlist state
     ▼
FastAPI + SQLAlchemy ─── SQLite or PostgreSQL
     │
     │ background refresh and shared cache
     ▼
  yfinance market data

The backend refreshes market data through a shared cache rather than fetching the same symbol independently for every request. The market-data layer, change-detection rules, API routes, and presentation layer are kept separate so the core behavior remains testable and explainable.

Every non-obvious choice behind this shape, and the real bugs found along the way, is written down with dates and reasoning in ENGINEERING_DECISIONS.md — 20+ entries covering why change detection is rule-based instead of ML, why an LLM layer was considered and declined, why polling is popularity-weighted, and production issues that were hit and fixed (a CORS misconfiguration, an incomplete NaN fix, a currency-column bug).

backend/
  app/
    routers/          auth, watchlists, charts, history, and related symbols
    services/         market data, change detection, and optional summaries
    models.py         database models
  tests/              backend behavior and rule tests
frontend/
  src/
    components/       dashboard, charts, drawers, watchlists, and navigation
    pages/             landing and authentication screens
    hooks/             watchlist state and refresh behavior
  public/              Drift logo and product artwork
PROJECT_BRIEF.md       product specification and rule rationale
ENGINEERING_DECISIONS.md architecture and scope decisions
vercel.json            production build configuration

Build the frontend with:

cd frontend
npm run build

121 backend tests across 6 files, all passing:

cd backend
python -m pytest -q

Built the solution that could be defended, not the one that looks complete from a distance:

• Single market-data source (yfinance). No fallback provider yet — a failed fetch is shown as visibly stale, never silently swapped for cached-as-current data, but there's no second source to fail over to. The data layer sits behind an interface specifically so one could be added without touching change-detection logic.
• One fixed benchmark (Nifty 50) for every symbol, not smart-matched to a stock's home exchange — a real simplification for a mixed US/India watchlist, disclosed rather than assumed correct.
• Polling is popularity-weighted, not load-tested at real scale. Each symbol is fetched once per interval and fanned out to every watcher, never once per user, but this hasn't been proven under production-scale concurrent load.
• Moving-average crossovers and any news/sentiment signal are deliberately out of scope (v2). MVP is price-move + volume-spike + 52-week-cross rules only; the remaining time went to resilience and edge cases instead of a fourth rule type.
• No automated accessibility or cross-browser testing. Manually checked, not covered by the test suite above.

Drift is an explainable watchlist and attention tool. It is not a broker, does not place trades, and does not tell users what to buy or sell. It helps investors understand what changed and decide what to investigate next.

• Live demo
• GitHub repository
• Product walkthrough
• Technical explanation
• Product brief
• Engineering decisions

This project is provided for demonstration and evaluation purposes.