The Build Story

1.2M+ lines.
One engineer. 18 months.

This is how the most comprehensive automotive AI platform was built from a blank file — and why that changes what you're buying.

Join the Waitlist → Architecture overview
By the numbers

What 1.2M+ lines looks like

1.2M+
Lines of original code
Zero outsourced, zero white-labeled
225
Public HTML pages
Every role, every module covered
720
JavaScript source files
In src/ — all production logic
239
PostgreSQL tables
21 departments, fully normalized
29
AI agents in production
27 live, 2 demo — all running nightly
18
Months to production
First line to live platform
A typical Series A startup has 50–100K lines of application code. LouieAuto has 4× that — built by one engineer, for one industry, over 18 months of field research.
The sequence

What got built, and in what order

The build order matters. The first thing built was not the dashboard — it was the brain. Everything else was built to serve it.

Months 1–3
The lender routing brain
The first feature built was the most important: a FICO/LTV/lender weight table that learns from funded outcomes. Not a form. Not a dashboard. The brain. Every other feature was designed to feed it data and surface its output.
Months 4–6
The desk
Four Square calculator, pencil, deal structure, taxes and tags, compliance layer. Everything needed to sit in a deal from first pencil to signed contract without touching paper or a legacy DMS.
Months 7–9
The BDC and CRM
Lead intake, routing queue, follow-up sequences, ADF parsing, attribution chain. The first time a lead could go from source to funded deal without touching a legacy tool. Every step logged for the closed-loop feedback.
Months 10–12
The agentic layer
8 agent teams (29 specialized agents), nightly cron jobs, voice AI. The platform stopped being a record system and became something that acts while you sleep. 1:00am brain learning cycle. 1:30am lender rebalancing. 7:00am morning briefing delivered. Every 15 minutes: service drive equity scan.
Months 13–15
PostgreSQL migration
239 tables. The flat SQLite prototype — fast for building — replaced with a schema designed to survive a PE firm's data room. 21 departments, fully normalized, with a migration script that can be audited line by line.
Months 16–18
The moat features
Louie Live voice AI, ElevenLabs TTS, the attribution model, the Exit Value Tracker. These are the features that would cost a competitor 2 years to build — because they require 18 months of field research to design correctly.
What was not outsourced

Every line is original

No white-labeled CRM dropped in. No third-party deal desk UI reskinned. No outsourced AI model wrapped with a thin layer of automotive UX.

The attribution methodology, the lender routing logic, the nightly agent architecture, the BHPH collections model — all built from scratch for this industry specifically. That means when you buy the platform, you own the original source. Not a license to use someone else's source. Not a reseller agreement. The code.

"You own the software. It runs on your server. If we disappear tomorrow, your platform keeps running. The brain keeps learning. The deals keep funding."
What this means if you're buying

The gap is 18 months

When you buy a platform built this way, you own something that can't be replicated by an incumbent with a month of engineering sprints.

The 239 tables took 18 months of field research to normalize correctly. The 8 agent teams — 29 specialized agents — took 12 months of iteration to make trustworthy. That time is gone. The gap between this platform and a competitor who starts today is 18 months minimum — and the gap grows every month that the brain learns from funded outcomes.

For a dealer: this means buying a platform that gets more accurate the longer you use it. The routing model improves from a 68% pre-deployment baseline to an observed 83% first-look accuracy (see /facts for methodology). A competitor who starts today starts back at that same 68% baseline.

For a strategic acquirer: the installed base doesn't just give you customers. It gives you trained models — a FICO/LTV/lender weight map built from that store's real funded deals. You acquire the learning, not just the interface.

See it yourself

The platform is live. The demo is real.

Every feature described on this page is running in production. The brain is learning. The agents are running. Walk through the demo and see what 18 months of field research looks like from the inside.

Join the Waitlist → Architecture overview