An automated intelligence subscription — "a sharp AI analyst in your inbox every morning" — for the sports value-betting niche. One research run per day → a branded brief → free-teaser + paid-full tiers over email + Telegram. Deployed, autonomous, paid billing live end-to-end.
🌐 Live: oddsedge.win · 📊 Track record: oddsedge.win/results 🔒 Engine + niche config: private (the research prompt + edge logic is the moat)
OddsEdge is a niche-agnostic intelligence-subscription engine I built and run autonomously on a DigitalOcean droplet. A scheduled job researches one niche once per day, Claude synthesizes a sharp branded brief, and it's delivered to subscribers over email + Telegram. The free tier is a 2-item teaser (the funnel); the paid tier is the full edition. One research run serves the entire list, so cost is flat (~a few cents of tokens per issue) while revenue scales with the subscriber count.
It's content, not money management — no custody or regulatory drag — and it deliberately reuses ~80% of the T BOT stack (cron → LLM → notify).
Status: deployed + autonomous; paid billing (Whop) confirmed end-to-end. Currently in the validation phase — proving demand before scaling the paid tier.
| Component | Role |
|---|---|
| Research engine | RSS + live-odds APIs → Claude → strict-JSON brief → Apple-light HTML (teaser vs full) → email + Telegram |
| Niche config | One JSON file per niche (brand, audience, sources, tiering, schedule, disclaimer) — niche #2 is a new config + one cron line |
| Paywall | A subscriber's tier (free → paid) is the entire gate; a payment webhook flips it |
| Landing + web | Cloudflare Pages — signup, results, full-edition sample, upgrade → Whop; KV-backed opt-in |
| Track record | Auto-settles published plays vs final scores → a public ledger (only graded plays show) |
| Reel pipeline | Claude designs the HTML; infra renders (VO + captions + headless Chrome + ffmpeg) + auto-publishes to YouTube |
| Ops reporting | Run / payment / lead / system events flow into the unified ops dashboard |
The engine knows nothing about sports. Every niche-specific detail — brand, audience, research prompt, data sources, tiering, schedule, disclaimer — lives in a single JSON config. Launching a second vertical is a new config file + one cron line, not a new codebase.
One research run per day serves every subscriber, free or paid. Token cost is ~$0.05–0.20/issue regardless of list size, so unit economics improve monotonically with the list. Break-even is roughly 15–30 paid subscribers.
Every brief emits machine-readable positions. A settlement job grades them against final scores and publishes a public ledger at /results — only graded plays appear, no cherry-picking. Transparency is the trust mechanism for a gambling-adjacent product.
There's no per-article entitlement system. A subscriber's tier (free → paid) is the paywall; a payment webhook flips it and the next brief renders the full edition instead of the teaser. Minimal moving parts, nothing to leak.
Positioned as sports data/analysis, not "guaranteed picks." An 18+ / responsible-gambling disclaimer is baked into every brief, and payment processing is chosen for the category.
Landing, signup, results, and the upgrade flow run on Cloudflare Pages + Functions + KV; billing is Whop (Standard Webhooks). The whole public surface is edge-served, while the research engine + niche config stay private.
DigitalOcean droplet (cron, autonomous)
┌──────────────────────────────────────────────────────────┐
│ daily_pipeline.sh (0 6 * * *, self-git-pulls) │
│ │
│ gather() ─→ RSS feeds + live-odds API │
│ research() ─→ Claude → strict-JSON brief + positions │
│ render() ─→ Apple-light HTML (teaser | full) │
│ send() ─→ email (Resend) + Telegram │
│ resolve() ─→ grade positions vs final scores │
│ reel() ─→ VO + captions + video → YouTube │
│ ops() ─→ run / payment / lead events → ops spine │
└──────────────────────────────────────────────────────────┘
│ │
▼ ▼
Cloudflare Pages (oddsedge.win) Unified Ops Dashboard
signup · results · upgrade→Whop (one feed across products)
Built solo. The engine + research prompt are private — the edge is the product; this repo is an architecture showcase.