Metals Trading Automation Suite

The Challenge

The team was manually monitoring multiple auction and store sources, tracking hundreds of product prices in spreadsheets, making gut-feel bidding decisions, and managing customer relationships across disconnected tools. In a market where auctions last 30 minutes and prices shift by the second, information delay meant lost margin — and important deals slipped through every week because no one could watch every screen at once.

We started with a single question: "Walk me through your day." What we found wasn't a tooling problem — it was an entire workflow held together by alt-tabbing, Excel files, and memory.

What We Built

1. Browser Extension — Automated Bidder

A browser extension that integrates directly into the auction platform the team already uses every day.

• Reveals hidden bid prices that the platform obscures, giving traders real-time competitive visibility
• Automated order placement — traders set limit prices, and the extension places orders within milliseconds of an auction opening
• Sub-second execution with retry logic and server-time synchronization to eliminate timing errors
• Multi-account support for managing orders across company entities
• Zero disruption — works inside the existing browser workflow, no new tools to learn

2. 24/7 Market Intelligence System

An event-driven daemon that monitors the entire market continuously without human intervention.

• Precise event scheduling — captures prices at exactly the right moments (1 second after auction open, 5 seconds before close) instead of wasteful polling
• Adaptive monitoring — normal intervals (5 min) tighten automatically as auctions approach close (10 sec)
• Inventory tracking — detects what sold, what moved between sources, and what got relisted across all market sources
• Spec sheet automation — downloads full product specifications via browser automation at key moments
• Historical database — every price point, every session, every inventory change stored and queryable

3. AI-Powered Analytics Dashboard

A web application where the team gets answers, not just data.

• Natural language queries in their own language — ask "which products had the biggest price swings this week?" and get a table back, no SQL required
• AI deal scoring — a custom ML model rates every product 0–100 based on historical pricing, specs, and market context, color-coded so traders spot opportunities at a glance
• Price history charts — track any product's price over 30/60/90 days across all sources
• Sold items analysis — filter by week, grade, finish, source. Export to CSV for margin analysis
• Session comparison — diff any two auction sessions to see what's new, removed, or repriced
• Package deal analytics — weighted average pricing for bundled blind-auction lots

4. Integrated CRM

A customer management system connected to the same pricing data.

• Account & contact management — customers, suppliers, contacts, with tagging and search
• Call recording → AI summary pipeline — upload a call, get back a transcript, summary, and action items linked to the right contact
• Living context — every contact page shows a brief of past interactions, what was promised, and what's still pending
• Inventory positions — track customer-specific inventory with bulk import
• CRM-aware AI queries — "which VIP accounts haven't been contacted in 30 days?" returns the list instantly

5. Production Infrastructure

• Docker Compose with four services on a single ARM cloud instance
• Cloud infrastructure provisioned with Terraform
• One-command deploy pipeline (local build → push → pull → restart)
• TLS termination via origin certificates
• Authenticated dashboard with JWT sessions

The Aha Moment

Once aggregated sold-by-spec-family data was visible for the first time, both we and the client started spotting trends neither of us expected. Specs the team had considered "hot" were actually sitting; quieter categories were moving fast. The conversation shifted from reactive bidding to strategic positioning — and the team began trading product categories they'd previously avoided because the cognitive load was too high.

Results

• Auction monitoring went from manual spreadsheet tracking to fully automated, 24/7
• Order placement went from racing to click buttons to sub-second automated execution
• Price analysis went from gut feel to quantified deal scores with historical context
• Customer management went from scattered notes to a unified system with AI summaries
• The team asks questions in their own language and gets data-backed answers in seconds
• Total infrastructure cost: about $50/month