Stock Screening & Technical Analysis Platform
A screening platform for listed equities combining candlestick pattern recognition with fundamental scoring, built on a resilient market-data pipeline that keeps working when the upstream data source rate-limits.
Discuss a Similar ProjectProject Details
Client
A market research publisher (confidential)
Industry
Fintech / Market Research
Technologies
Project Overview
The client publishes market research and wanted to give readers the screening tools behind it, rather than only the conclusions. That meant two engines: technical pattern detection across the daily price history of the listed universe, and fundamental scoring using established published methodologies.
The harder half of the work was not the screening logic but the data. Free market-data sources rate-limit aggressively and fail intermittently, so the pipeline had to treat upstream failure as normal rather than exceptional.
The Challenge
Detecting technical patterns across the full listed universe means fetching and maintaining daily price history for thousands of symbols.
The upstream market-data source rate-limits and returns intermittent errors, so naive sequential fetching neither completes nor recovers.
Fundamental scoring methodologies each need a different set of financial statement inputs, with incomplete data handled rather than crashing the scan.
Bulk data refreshes run for long enough that a user needs live progress, not a request that appears to hang.
Refresh jobs must not overlap — a second run starting mid-refresh would corrupt the working set.
What We Built
Candlestick pattern recognition covering 34 distinct patterns, browsable by category — bullish and bearish, continuation and reversal, plus neutral.
Multi-pattern combination detection, so co-occurring signals surface rather than being read one at a time.
Three fundamental scanners implementing published methodologies: Piotroski F-Score, Peter Lynch criteria and the Greenblatt Magic Formula.
A database-first ingestion pipeline with the upstream API as fallback, using concurrent fetching with exponential-backoff retry and explicit rate-limit handling.
Scheduled data refresh guarded by a lock, with retry limits, health checks and a recorded data-quality score per run.
Live progress streamed over websockets during bulk ingestion, so long refreshes report status instead of appearing stalled.
Per-symbol detail views with multi-timeframe charting.
An editorial section with categories and image handling, so research content sits alongside the tools.
Delivered
Readers moved from consuming the firm’s conclusions to running the screens themselves — the research tooling became the product rather than just its output.
Screening covers the listed universe: 34 candlestick patterns across bullish, bearish, continuation, reversal and neutral categories, plus detection of co-occurring signals rather than one pattern at a time.
Three published fundamental methodologies — Piotroski F-Score, Peter Lynch criteria and the Greenblatt Magic Formula — each usable as a standalone scanner.
Upstream rate limits stopped being an outage: the pipeline treats vendor failure as routine, reading database-first with concurrent fetching and exponential-backoff retry behind it.
Long refreshes report live progress over websockets instead of appearing to hang, and a lock prevents a second run from corrupting the working set mid-refresh.
Data staleness is observable rather than assumed — every scheduled refresh records a quality score.
57 routes across six functional modules, mixing server-rendered pages with JSON endpoints.
Building something similar?
Tell us what you are trying to ship and we will map out scope, timeline and a transparent estimate.
Book a Strategy Call