An autonomous agent pipeline that turns a folder of trucking company CSVs into a complete sales demo — insights report, company profile, and a working ops platform — with no human steps between input and output.
Carrier Demo Factory is an autonomous agent pipeline that turns a folder of trucking company CSVs into a complete, deployed sales demo — with no human steps between input and output. One command triggers the Claude Code pipeline: it analyzes the data like a consultant, produces three polished deliverables, and deploys them to a live VPS.
This is a prototype, not a finished product. It ships with a demo data generator (Phase 0) that creates fictional carriers with synthetic CSVs — useful when you don’t have a real client’s data and want to showcase the pipeline anyway. For a real engagement, skip Phase 0 entirely: drop in the carrier’s actual CSV exports and the rest of the pipeline runs identically. The project was built to answer a real sales question: how do you walk into a cold meeting and immediately demonstrate that you understand a carrier’s operation better than they do? The answer is to show up with a working product already built on their numbers.
companies.json state file updated by deploy scripts so the portfolio page never requires a manual editcurrent_state.txt as a simple checkpoint so a failed phase can be retried without rerunning the entire pipelinerun-local.sh and run-cycle.sh scripts with identical logic, only differing in whether deploy steps are executedA self-contained HTML file with Plotly charts covering every major pain point. No login required — built to be emailed to the prospect, their accountant, or their bank before the pitch even begins.
A public-facing profile that tells the company's story, surfaces key stats, and frames pain points in plain language. Designed to create recognition — the prospect feels understood before they see a product.
A multi-page Streamlit application deployed over HTTPS and running on the prospect's actual data. Not a mockup — a working product accessible from any device before the sales call starts.
Claude writes its own Python analysis scripts and follows threads in the data independently. Every demo surfaces different pain points because every dataset is different — nothing is hard-coded.
The agent registers systemd services, writes Caddy reverse proxy routes, serves everything over HTTPS, and updates the portfolio JSON — no manual infrastructure steps after the pipeline runs.
The pipeline generates demos for owner-operators, small fleets, regional carriers, and mid-size carriers — each with carrier-appropriate data volumes, pain patterns, and platform feature sets.
The core pipeline is phases 1–4, driven by a bash state machine (run-cycle.sh) that advances through each phase invoking Claude Code with a scoped prompt. An optional Phase 0 exists for demo purposes when no real carrier data is available:
company_profile.md and the raw CSVs, writes and runs Python scripts to explore the data, and produces a structured list of pain points with exact figuresdeploy-app.sh to register and start itEach phase reads outputs from the previous phase and writes its own outputs to a company directory. The state machine tracks the current phase in current_state.txt — if a phase fails, the pipeline restarts from that phase, not from the beginning.