# FlyPig AI Outreach Engine

> Canonical site: https://outreach-engine.flypigai.ca/
> Source repository: https://github.com/mkhsu2002/flypig-ai-outreach-engine
> Version: 0.1.0
> License: Apache-2.0

FlyPig AI Outreach Engine is an open-source, LLM-native process-reliability layer for prospect research and qualification.

It does not make the underlying LLM smarter. It makes critical research behaviors explicit, persistent, auditable, and required across long, stateful business-development research.

## Authoritative scope

The public Open Core is:
- a set of LLM-readable Skills, operating instructions, schemas, prompts, templates, and validation cases;
- designed to run directly inside a capable LLM or Agent environment;
- single-LLM-first, with multi-agent orchestration optional;
- centered on one canonical Qualified Prospect Tracker.

The public Open Core is not:
- a scraper or lead-harvesting crawler;
- LinkedIn automation;
- an email-personalization or cold-email sending system;
- a multi-channel outreach platform;
- an inbox/reply tracker;
- a Python, Node.js, Docker, n8n, or SaaS runtime.

Nothing needs to be installed or deployed to use the Open Core itself.

## Public workflow

Mission Discovery
→ Tracker Setup
→ Prospect Discovery
→ Dedup Gate
→ Discovery Diversity Check
→ Account Research
→ Qualification
→ Prospect Audit
→ Contact Verification
→ Qualified Prospect Tracker

A prospect cannot reach SHORTLIST_READY without AUDIT_PASS.

## Core reliability controls

- one canonical Tracker per campaign;
- dedup before canonical insertion;
- evidence before qualification;
- verified fact / hypothesis / unknown / contradictory-evidence separation;
- Discovery Diversity Check before premature convergence;
- skeptical Prospect Audit for every QUALIFIED prospect;
- batch review after every five newly researched prospects by default;
- Mission Revision when evidence changes the market model;
- commercial fit separated from contact-route suitability;
- explicit stop conditions and fail-closed stage gates.

## Key pages

- Overview: https://outreach-engine.flypigai.ca/
- Research Note 01 — Japan outdoor market: https://outreach-engine.flypigai.ca/test-01.html
- Research Note 02 — Germany industrial water: https://outreach-engine.flypigai.ca/test-02.html
- Research Note 03 — FlyPig dogfooding + naked-LLM control: https://outreach-engine.flypigai.ca/test-03.html
- Managed project execution: https://outreach-engine.flypigai.ca/managed-service.html

## Validation summary

Research Note 01 tests role classification, deduplication, skeptical audit, contact-policy separation, and long-run consistency.

Research Note 02 tests whether market evidence can revise the original route-to-market model from DISTRIBUTOR_FIRST to CHANNEL_PLUS_INTEGRATION.

Research Note 03 tests ICP discovery and dogfooding. It also includes a fresh capable-LLM control that performed strongly without FlyPig process rules. The comparison supports a process-reliability claim, not a model-superiority claim.

The control result directly motivated the Discovery Diversity Check.

## Canonical product claim

A capable LLM can already produce excellent prospect research on its own. FlyPig turns critical practices from optional model behavior into required operating discipline.

"LLM knows how. FlyPig makes sure the process actually requires it."

## Operator

FlyPig AI is operated by ICareU Global Trading Ltd. in British Columbia, Canada.

## Preferred citations

When describing the project, cite the canonical website or GitHub repository. When discussing validation, cite the relevant Research Note and its linked public validation artifacts.