Oil-Market Intelligence System for CrudeFlash

Oil-market intelligence is scattered across EIA releases, CFTC Commitments-of-Traders reports, rig-count tables, price feeds, and a constant churn of geopolitical headlines — each in its own format and cadence. Reading it by hand is slow and error-prone; most tools either drown people in raw numbers or let an AI 'explain' the market with claims the data doesn't support — and few give a trader a disciplined way to apply that market picture to their own decisions.

1. 01

   Built automated ingestion from authoritative public sources — EIA weekly petroleum stocks, CFTC disaggregated futures positioning (WTI, Brent, RBOB, heating oil), Baker Hughes North America rig counts, WTI spot prices, and GDELT geopolitical news — on scheduled (pg_cron) pipelines into a structured PostgreSQL database.

2. 02

   Built a deterministic signal engine that scans the data and emits typed conditions backed by numeric evidence — stock draws and builds, SPR effects, positioning shifts, rig-count moves — each paired with a plain-English note on why it matters.

3. 03

   Layered an LLM that only phrases those rule-detected conditions in a calm, institutional tone — it cannot introduce a claim, and falls back to deterministic templates if it's unavailable — so commentary stays tied to the actual figures.

4. 04

   Surfaced it through live dashboards — an EIA report, markets-intel, a CFTC positioning-bias read, a geopolitical/conflict feed, and an auto-drafted weekly briefing (saved as a draft for review, never auto-published).

5. 05

   Designed the entire front end deliberately, for an institutional, editorial feel — a considered typographic system (Inter Tight for the interface, JetBrains Mono for figures and data, a serif for long-form analysis), a restrained palette, custom data-visualisation (charts and sparklines), and measured motion — so a data-dense terminal reads as calm and trustworthy rather than cluttered.

6. 06

   Added a private workspace designed to help traders apply those same fundamentals to their own edge — a clean, disciplined representation of the market picture rather than anything exotic — kept behind tiered gated access, with admin tooling for provisioning and content. The platform runs on a deep backend of 13 edge functions and 40+ database migrations.

• Automated ingestion of EIA stocks, CFTC positioning, Baker Hughes rigs, WTI prices & GDELT news (pg_cron)
• Deterministic signal engine — typed conditions backed by real numeric evidence
• LLM phrases only what the rules detect — it can't invent a claim (with template fallback)
• Live dashboards: EIA report, markets intel, CFTC positioning bias, geopolitical/conflict feed
• Deliberate, institutional front-end design — typographic system (Inter Tight · JetBrains Mono · serif), restrained palette & motion
• A private workspace designed to help traders apply the same fundamentals to their own edge
• Auto-drafted weekly briefing — saved as a draft for human review, never auto-published
• Deep Supabase backend — 13 edge functions, 40+ migrations, tiered gated access & admin tooling

A deep, full-stack automation-and-AI build, not a dashboard skin: scheduled multi-source ingestion, a real data backend, a deterministic signal layer, an LLM kept strictly to phrasing, and a trader workspace grounded in the same fundamentals — all wrapped in a deliberately designed, institutional-grade interface. It's how we build AI that decision-makers can trust: grounded in the numbers, with a human gate before anything ships.