AI Agent for Compressor Manuals | Compressor World

Compressor World is a Massachusetts online store for industrial air compressors, operating since 2000. Brocoders built AskCW.ai for it, an AI assistant that answers technical and selection questions from the company’s own manuals and spec sheets. Every fact in an answer cites its source file and page. Buyers get 5 free queries a month.
Launch a product
Compressor World needed accurate, document-based answers to technical buyer questions around the clock, instead of relying only on phone, email and live chat. It also needed a way to promote partner brands through conversational advertising, a free tier and a paid plan.
The agency's contribution
Brocoders handled AI product development end to end, covering both the front end and the back end. That included discovery at the April 8, 2025 kick-off, UI design, the document ingestion and retrieval pipeline, the chat and accounts, and the Phase 2 monetization features. The team was a project manager, a front-end engineer, 2 back-end engineers, a designer, a DevOps engineer and, in phase 2, a QA engineer.
What was delivered
By what was delivered and live in production: 4,090 documents indexed from 3 sources 2 build phases delivered in 2025 A live product with accounts, free and paid plans, sponsored cards and follow-up suggestions Verified answer behavior: source file and page, “not found” replies and the accuracy disclaimer
Measurement period: Phase 1 from April 14, 2025, phase 2 from July 31, 2025, with the document count as of July 2025 and the live product checked on October 16, 2026.
Context and limitations: No usage, conversion or revenue figures are disclosed. The retrieval layer can’t read images directly, so for diagrams the model works from the figure descriptions in the documents. The chat carries a note that answers may not be 100% accurate. The Pro price is set by Compressor World. AskCW.ai runs as its own website, and widget embedding was only being prepared in May 2025.

Challenge
Compressor World wanted a domain-specific AI agent for its business, with accurate, document-based answers around the clock.
The main challenges were:
Technical knowledge was spread across thousands of manuals, spec sheets and FAQs in Google Drive, Dropbox and Google Sheets. Buyers (about 80% technical, 20% non-technical) describe a job in plain words, not in specs. Answers had to be trustworthy: no invented specs, no prices from the AI. The company wanted a business model: a free tier, a paid plan and sponsored product placements. Diagrams in the documents could not be read directly by the retrieval layer.
Solution
Brocoders built AskCW.ai as a retrieval-augmented generation (RAG) assistant in two phases in 2025.
The solution included:
A scheduled ingestion pipeline through Brocoders’ Bridge knowledge base, copying files to AWS S3 and indexing PDF, CSV and XLSX files. Retrieval with LlamaIndex and answer generation with OpenAI GPT-4o, using figure descriptions for diagrams. An answer format that turns a buyer’s job description into requirements and selection criteria, then suggests options with the file and page under each fact. Clear fallbacks: “not found in the documentation” for missing specs, and a handoff to sales for price questions. A chat interface, accounts, query limits and an admin panel, built with Next.js, NestJS and PostgreSQL. Phase 2 monetization: a Free plan (5 queries a month), a Pro plan (unlimited access), marked “SPONSORED” product cards inside answers and follow-up suggestions such as “Talk to an expert”. UI design in Figma.
Results
4,090 manuals, spec sheets and FAQs indexed from 3 document sources. Two build phases delivered in 2025, starting April 14 and July 31. AskCW.ai is live at askcw.ai with accounts, free and paid plans, sponsored cards and follow-ups. Answers are traceable to a source file and page, with a “not found” reply when a value isn’t in the documents. The knowledge base stays current through scheduled re-indexing, so changed spec sheets reach the assistant without manual uploads.
Project Details
2025


