Project Overview
The Problem We Saw
Industrial equipment markets are deeply inefficient. Sellers list assets across dozens of marketplaces. Buyers cannot monitor all of them. Good equipment sits undervalued while buyers pay more elsewhere.
The result is a persistent information gap. A piece of equipment worth $80,000 may sell for $30,000 at auction because the right buyer never saw it. The buyer who needed it paid $90,000 from a dealer. Both sides lost.
We built a system that closes that gap.
What We Built
A ten-stage pipeline from discovery to revenue, with humans in the loop at every decision point.
1. Discover
Automated connectors scan 13 industrial marketplaces every 4 to 6 hours, pulling listings for heavy equipment, machinery, and industrial assets.
2. Normalize
Listings are deduplicated, categorized, and enriched with specifications, photos, and location data. A single asset appears once, regardless of how many sites list it.
3. Score
Each asset receives a category-aware deal score: margin potential, buyer match probability, transaction speed, and data confidence. The system ranks opportunities, not just listings.
4. Match
Scored assets are matched against a verified buyer database using eight criteria, including category preference, geography, budget range, and purchase history.
5. Review
A human deal desk reviews every scored opportunity before any action is taken. The system recommends. Humans decide.
6. Coordinate
For approved deals, the system coordinates inspection, negotiation, and transaction through trusted platforms. We never hold buyer funds. Transactions flow through licensed marketplaces or escrow services.
System Architecture
Eight independent engines, each with a single responsibility, connected by a common data contract.
E1
Supply Discovery
Finds undervalued assets across 13 data sources, including auction sites, classifieds, and government surplus platforms.
E2
Buyer Intelligence
Builds buyer demand profiles with intent scoring. Buyers are validated and tiered using a seven-factor formula covering domain verification, business legitimacy, and purchase history.
E3
Market Intelligence
Valuation engine that pulls comparable sales data and computes estimated market value with category-aware multipliers.
E4
Buyer-Asset Matching
Scores assets against the buyer database using eight matching criteria to surface the right opportunity for the right buyer.
E5
Opportunity Packaging
Packages scored deals as full opportunity briefs with economic analysis, risk assessment, and recommended action.
E6
Human Deal Desk
Human-in-the-loop workflow with approval gates. Every deal beyond screening requires explicit human approval before any outreach or action.
E7
Transaction Coordination
Coordinates inspection, payment, shipping, and documentation through trusted third-party platforms. No fund custody.
E8
Revenue Engine
Tracks sourcing fees, invoicing, and payment status on closed transactions.
Outcomes So Far
The system is operational. Here is what it has produced.
555
Assets Ingested
Across 13 marketplaces
8
Verified Buyers
Tiered A through D
13
Data Connectors
Built and tested
68
Tests Passing
Full backend coverage
Supply side validated
13 connectors built across auction sites, classifieds, and government surplus. 555 real assets ingested, deduplicated, and scored. The supply pipeline works.
Demand side seeded
8 verified buyers in the database with domain validation, tier scoring, and intent profiles. The matching engine has produced 11 buyer-asset matches.
Compliance framework
Every source, contact, and deal carries a compliance state (GREEN, YELLOW, RED). No automated action against restricted sources. Five layers of safeguards protect buyer and seller information.
Human-in-the-loop proven
The deal desk workflow is operational. Assets flow through review queues with approve or pass decisions. The system surfaces opportunities; humans make the call.
Design Principles
Six principles that shape every decision in the system.
AI ranks, humans decide. The system surfaces and scores opportunities. A human approves every buy or pass decision.
Compliance first. Every source, contact, and deal has a compliance state. No automated action against restricted sources.
Intent-based, not list-based. The system detects actual buying and selling behavior, not just listings.
No fund custody. Transactions flow through existing licensed infrastructure. We never hold buyer funds.
Lean infrastructure. Open-source everything. Pilot operates on less than $85 per month.
Evidence quality outranks model sophistication. A simple score on reliable signals beats an advanced model on weak data.
Technology Stack
Built with a modern, async-first, type-safe stack.
Backend
Python 3.12 + FastAPI with async SQLAlchemy, Pydantic schemas, and auto-documented OpenAPI endpoints.
Frontend
React 18 + Vite + TypeScript with a custom token-driven design system supporting dark and light themes.
Data Layer
PostgreSQL with geospatial extensions, Redis for async task queuing, and Qdrant for vector-based similarity matching.
Browser Automation
Playwright for multi-browser scraping with stealth mode. Managed scraping via Apify actors for JavaScript-heavy sites.
Storage
Backblaze B2 (S3-compatible) for photo storage. Assets hosted on controlled infrastructure, not hotlinked from source sites.
Infrastructure
Docker Compose orchestrating six services. CI/CD via GitHub Actions. Designed for hybrid deployment across serverless and persistent hosts.
Plan Moving Forward
Where This Goes Next
The engineering is complete. The next phase is operational.
01
Scale data connectors. Unblock remaining sources and expand coverage from 7 active to all 13 marketplaces.
02
Grow the buyer network. Expand from 8 to 20+ verified buyers. More buyers means more matches, more deals, more data for the scoring engine.
03
First closed transaction. Move from matched opportunities to a completed deal with sourcing fee collected. Prove the full pipeline end to end.
04
Production deployment. Move from Docker development to production hosting with automated monitoring and daily data quality reports.