Shared fault models for rooftop solar installers, without pooling their data
TL;DR Faultpool gives small rooftop solar installers early warning of inverter and string faults across the homes they maintain. Each installer's data stays on its own server, but the fault model learns from all of them together. Installers pay per site per year, because fewer truck rolls and fewer angry customers protect their margin.
India had 33.41 GW of grid-connected rooftop solar by the end of September 2026[1], a good chunk of it on homes under PM Surya Ghar, which targets about one crore households and pays subsidies of ₹30,000–78,000 per home through registered vendors[2]. Those vendors are mostly small firms, and many also sell annual maintenance contracts[3].
A small installer looks after a few hundred systems. It sees too few inverter failures to learn the patterns that come before them, so it finds out when a customer calls about a high bill. Inverters are known to cause a large share of solar system faults, though the one field study I found gives no figure[4], and there's no public failure rate for Indian rooftop systems that I could find.
Faultpool installs a small agent on each installer's server or cloud account. It reads the generation and alarm data the inverters already send and compares each site with its expected output for that day's weather. The fault model trains through federated learning - each installer's copy learns locally, and only model updates travel to Faultpool, never the raw readings or customer names. A 2026 study of substation equipment found this approach workable across utilities with different data, with F-scores of 0.60–0.88, though on simulated data[5].
Each morning the installer gets a ranked list of sites losing output, inverters showing early fault patterns and the likely cause.
I reframed the seed. It pitched federated learning for critical infrastructure such as utilities. One state utility has few peers willing to federate, and it can share data by contract anyway. Federated learning fits better where many small competitors own similar assets and refuse to share customer data.
The installer pays and the homeowner benefits. My working assumption is ₹300–800 per site a year, which is small next to a system that costs about ₹50,000 per kW to build[3]. Inverter brands and lenders who finance rooftop systems are a second buyer later. The pitch to an installer is simple - one avoided wasted visit a month should pay for a few hundred sites.
PM Surya Ghar has put large numbers of homes in the care of small installers within two years, and open-source federated learning tools such as Flower, with 180+ contributors, now make the training plumbing cheap[6]. Installers who sell annual maintenance contracts also carry the cost of every missed fault[3].
Nobody publishes a figure for spending on rooftop solar monitoring in India. The firm number is the installed base of 33.41 GW of rooftop capacity[1] and a scheme aimed at about one crore homes[2].
If you want a real number then count the sites your first 20 installers maintain and price from that.
Inverter makers run their own monitoring clouds, but each sees only its own brand and rarely predicts faults. Larger installers run in-house operations teams[3]. Utility-grade asset platforms target big plants.
I found no product that pools fault learning across small Indian installers while keeping their data apart. That's the gap.
| Difficulty | Moderate (48/100) |
| Starting capital | ₹1.2–2.2 Cr |
| Time to launch | 9–12 months |
Reading data from many inverter brands, each with its own format, is slow and unglamorous work. Federated training across installers with uneven data is still research-grade. And the product only earns trust once it predicts a failure that a technician then actually finds on the roof, and does it more than once.
The software itself needs no sector licence. The rules that matter sit with your customers and their equipment.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
Don't trust a flag until a technician has confirmed it on the roof, and never let the system send commands to inverters.
Want to work out the AI layer for Faultpool? Talk to Yash →
Here's what can go south -
Good for a founder with solar operations or inverter firmware experience. If you want to sell federated learning as the headline then avoid it, because installers buy fewer site visits and not buzzwords.
Prove one thing first - across five installers, at least 60% of Faultpool's flagged sites show a real fault on inspection.
Plan for roughly ₹1.2–2.2 Cr, mostly on engineers, inverter integrations and the research work federated training needs.
No sector licence for the software. You need company registration, GST, DPDP processor contracts and a careful read of PM Surya Ghar vendor terms and inverter warranties.
Yes, any market with many small rooftop installers fits the model. Check local data protection law and the inverter makers' terms there first.
The free Faultpool validation kit has the line-by-line capital breakdown, the effort scoring, the data and contract checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for installers, and 5 assumptions to test. Download the kit →
Thinking about building Faultpool or something close to it? The make-or-break work is getting usable data out of a dozen inverter brands and proving early catches on real roofs, and that's the kind of plan I help founders put together. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.