
AI-Powered HVAC Label Recognition and Predictive Maintenance Platform
Challenge
Our client, an American startup, built an IoT HVAC product: smart devices that automatically clean A/C drain lines, plus web and mobile apps for B2C and B2B users to install, monitor and manage the devices and order replacement cartridges.
Moving into B2B, the client saw a gap: HVAC integrators arriving on site rarely know the exact air handler model, its warranty or service life, which makes preventive maintenance harder and increases costs and downtime. The new feature had to let users photograph the air handler label during setup and automatically extract brand, model, serial and product numbers, manufacturing date, tonnage, electrical details, horsepower, SEER rating, warranty status and filter configuration.
With 15+ major manufacturers and dozens of models each, a manually maintained reference database was not sustainable, so AI was the only viable path, and recognition had to stay cost-efficient at scale.
Our client, an American startup, built an IoT HVAC product: smart devices that automatically clean A/C drain lines, plus web and mobile apps for B2C and B2B users to install, monitor and manage the devices and order replacement cartridges.
Moving into B2B, the client saw a gap: HVAC integrators arriving on site rarely know the exact air handler model, its warranty or service life, which makes preventive maintenance harder and increases costs and downtime. The new feature had to let users photograph the air handler label during setup and automatically extract brand, model, serial and product numbers, manufacturing date, tonnage, electrical details, horsepower, SEER rating, warranty status and filter configuration.
With 15+ major manufacturers and dozens of models each, a manually maintained reference database was not sustainable, so AI was the only viable path, and recognition had to stay cost-efficient at scale.
Solution
WebbyLab designed and built a multi-layered AI label recognition pipeline. Users upload a nameplate photo from the mobile or web app; the backend queues a task in Google Cloud Tasks for stable asynchronous processing; an LLM performs the initial analysis and extracts key fields such as brand, model and serial number.
If the model is already known, details come from a cache instead of re-analysis. For new models the data is enriched with Vertex AI grounding over a curated library of manuals, PDFs and photos, and, for hard cases, with Google Search grounding. The system then returns structured JSON and normalizes it: filters invalid or duplicate photos, standardizes manufacturer names, decodes manufacturing dates from serial numbers using brand-specific rules, and calculates warranty status and estimated expiry before saving to the IoT HVAC monitoring platform database.
Stack: React Native, React.js, Node.js, ESP-IDF and Google Cloud Platform.
WebbyLab designed and built a multi-layered AI label recognition pipeline. Users upload a nameplate photo from the mobile or web app; the backend queues a task in Google Cloud Tasks for stable asynchronous processing; an LLM performs the initial analysis and extracts key fields such as brand, model and serial number.
If the model is already known, details come from a cache instead of re-analysis. For new models the data is enriched with Vertex AI grounding over a curated library of manuals, PDFs and photos, and, for hard cases, with Google Search grounding. The system then returns structured JSON and normalizes it: filters invalid or duplicate photos, standardizes manufacturer names, decodes manufacturing dates from serial numbers using brand-specific rules, and calculates warranty status and estimated expiry before saving to the IoT HVAC monitoring platform database.
Stack: React Native, React.js, Node.js, ESP-IDF and Google Cloud Platform.
Results
B2B integrators now get actionable information about HVAC units before arriving on site; end users get clarity on warranty status and recommended maintenance; the client gained a new revenue stream, a competitive edge and cost optimization.
The three-tier strategy balances accuracy and cost: cache retrieval costs about $0.09–0.28 per 1,000 queries, Vertex AI grounding about $2.59–7.78, and Google Search grounding, used only for rare or new models, about $35–105. Common queries hit the cheapest layer, so the system scales cost-efficiently while keeping high accuracy for edge cases. A team of 4 delivered the work in 4 months.
B2B integrators now get actionable information about HVAC units before arriving on site; end users get clarity on warranty status and recommended maintenance; the client gained a new revenue stream, a competitive edge and cost optimization.
The three-tier strategy balances accuracy and cost: cache retrieval costs about $0.09–0.28 per 1,000 queries, Vertex AI grounding about $2.59–7.78, and Google Search grounding, used only for rare or new models, about $35–105. Common queries hit the cheapest layer, so the system scales cost-efficiently while keeping high accuracy for edge cases. A team of 4 delivered the work in 4 months.