// jays.website / architecture

How It's Built

Every system on this site is production software running against real data on real infrastructure — no demos, no mockups. This page is the engineering tour: where the data comes from, how it's processed, and what keeps it running unattended.

The Big Picture

Linux data plant→ nightly Python pipelines→ SQLite core DB→ scanners & scoring→ static HTML / JSON build→ self-hosted nginx + Flask→ Cloudflare Tunnel→ Cloudflare edge

~80 scheduled jobs orchestrate everything. No CMS, no frontend framework — hand-built static pages fed by Python pipelines, rebuilt and redeployed automatically every market day.

The Systems

Trading Intelligence

Signals · Valuation · Regime · Delivery

Data In

Polygon.io grouped daily bars for a 2,212-symbol universe, loaded nightly into a central SQLite database
SEC EDGAR fundamentals for DCF valuation
Institutional 13F holdings for accumulation signals

Processing

12 automated scanners at market close — breakout, range compression, institutional, RSI staircase, ATR momentum, confluence, and more
Market-regime gate (breadth + trend) suppresses signals in hostile tape
Conviction scoring weights each scanner by its measured historical hit rate
Claude-based analyst ranks a top-25 nightly

Out

Live dashboard, per-scanner HTML reports, JSON APIs — all static, all auto-deployed
Telegram alerts for high-conviction setups
Public paper portfolio tracking every pick's outcome

2,212

Symbols

Scanners

264K+

Signals tracked

Daily

Auto-deploy

Full build notes: Trading →

Rental Intelligence

Parcels · Scoring · Ownership Graph

Data In

MassGIS statewide parcel data — every one of 305,476 multifamily buildings in Massachusetts
HUD Fair Market Rents (FY2025) by bedroom count and county
Town assessor records: units, year built, assessed value

Processing

Composite 0–100 deal score: yield proxy vs. HUD benchmarks, distress and ownership signals, data-quality confidence
Ownership resolution maps 30,000+ unique owners — portfolio holders, absentee and institutional-scale owners flagged

Out

County opportunity rankings and top-scored buildings
Interactive statewide map, filterable to street level
Free to explore — Pro tier (alerts, CSV export, cap-rate calculator) in the works

305,476

Buildings

14,982

Active scores

30k+

Owners mapped

Counties

Full build notes: Rental →

Business Intelligence

Spatial · Sector Gaps · Productization

Data In

OpenStreetMap business data across Massachusetts ZIP codes, towns, and counties

Processing

Industry density, diversity, and sector-gap analysis per zone
"Curiosity zones": areas where services are systematically missing relative to comparable zones

Out

Zone explorer, sector breakdowns, town-level drill-down
Proof of productization: a live niche directory built from the same pipeline — 1,445 companies across 20 city pages

Full build notes: Business Intel →

Infrastructure & Practices

Where it runs

Self-hosted Linux workstation as the data plant — pipelines, database, scanners
The same workstation serves the site: Cloudflare Tunnel → nginx → Flask, behind Cloudflare's edge
Oracle Cloud keeps an offsite mirror for backup; Tailscale private mesh connects machines; services run under systemd

Engineering choices

SQLite as the core store — zero-ops, fast, right-sized for the workload
Static-first delivery: pages are pre-built, so the site stays fast and cheap
~80 cron jobs with logging, log rotation, and a backed-up crontab

Operating principles

Measure outcomes before adding features — every scanner's accuracy is tracked against real results
Regime gating: don't emit signals the market context doesn't support
Dry-run flags on anything destructive; secrets live in env files, never in code

Stack

PythonSQLitepandasFlasknginxCloudflareOracle CloudTailscalesystemdPolygon.ioSEC EDGARMassGISHUD FMROpenStreetMapTelegram APIClaude API

Built and operated by one person. Want the longer story, or the person behind it?

About → LinkedIn