Blackbird

A Blackbird sees clearly from above, even in the dark.

Stats for Nerds:

  • Team Size: 1
  • Timeline: Seed stage (started May 26, 2026)
  • Language: TypeScript
  • Framework(s):
    • Supabase + Postgres
    • Edge / serverless runtime (scales to zero)
  • Tools:
    • Claude
    • Apollo + skip-trace + public records + AI web crawl (data sources)
  • Deployment: Not yet shipped
  • Repository: https://github.com/Serrowxd/Blackbird

The Concept

What inspired you?

Lead enrichment is fragmented and expensive when you do it naively. Single providers cap out at 40–60% match rates, no one provider is good at everything (email vs. mobile vs. firmographics vs. tech stack), and most teams either overpay for a full-suite tool or under-enrich and burn outreach budget on contacts that were never qualified. Meanwhile the data an AI can read — website copy, job postings, recent funding, product launches — gets ignored entirely by database-only approaches.

Blackbird is a single endpoint you drop into a workflow. Feed it a thin input — a name and domain, an email, an address — and it returns a structured, confidence-scored profile compiled from multiple sources, ready to route into outreach or a CRM. The endpoint is the product.

The Niche

Where it actually plays

Blackbird is the real-estate purchase-intent endpoint — one API call that takes a thin input and returns a scored, summarized prospect profile compiled from real-estate-relevant sources, with no PII persistence.

The existing real-estate intelligence tools — PropStream, BatchLeads, ListSource, PropertyRadar — are batch list-building products tied to a CSV download and a login UI. None of them are designed to be dropped into a Make.com or n8n workflow as a single per-call endpoint. Wholesalers, agents, and investors are duct-taping CSV exports through Zapier to fake exactly that. Blackbird skips the UI and goes straight to a real-time per-call API.

The agent-callable angle compounds it: the same properties that serve a human-workflow buyer — predictable schema, confidence scores, source attribution, idempotency, cost transparency, no PII storage — are exactly what an AI agent needs to call a tool reliably. That slot is still open.

The Design Philosophy

The load-bearing decisions

Waterfall-first. Multiple sources, sequential, per-field. Stop when a field is filled. Never rely on one provider.

Qualify before you enrich. Run firmographics first and gate expensive contact enrichment behind a qualification score, so unqualified leads never burn credits.

Per-field confidence. Every field carries a float from 0 to 1 and a source attribution. Downstream logic decides what to trust.

AI-augmented. Use LLM scraping for the unstructured signals no database has, and write a tuned “why now” summary on top.

Graceful degradation + cache-first. A failed provider gets logged and the pipeline continues; previously enriched records serve from cache with a field-level TTL, because enrichment credits are money.

The Ephemeral Architecture

Why "we never see your leads" is a feature

Blackbird doesn’t persist the lead PII. The database holds only config, a hashed audit trail, and analytics aggregates — never the lead data itself. That one decision does three jobs at once: it keeps Blackbird out of data-controller territory and shrinks the compliance surface; it earns trust with wholesalers and agents who treat their lead lists as paranoid currency; and it scales to zero with simple unit economics on the lead-data path.

Status

Seed stage. The enrichment pattern is researched, the positioning and runtime architecture are locked, and an accuracy benchmark is the next concrete step. The bet is robust + focused + acquirable — accuracy is the wedge, not a moat.