Naim Makhlouta
← All work
ProductionMay to June 2026Monitoring and upsell for solar installations

Solar Analytics & Upsell Agent

Turns raw solar monitoring data into reports and justified product recommendations for each client.

80+solar installations in the pipeline
2monitoring APIs unified
3report types generated automatically

The need

There was no central tool to monitor the performance of the installations. Raw data went from API to files and was then reviewed by hand, installation by installation, with no agreed criteria:

  • Slow. Each installation needed several tables and charts to be read closely.
  • Inconsistent. Conclusions changed with the analyst and their experience.
  • Unprioritised. Commercial opportunities were spotted case by case, and were often missed or found too late.

What I owned

The data pipeline and the AI agent, including working with technicians to turn their judgment into explicit decision rules.

The solution

A pipeline pulls data from both monitoring APIs, cleans and normalises it, and generates three kinds of report. An AI agent then reads the reports, diagnoses each installation and produces a structured table of recommendations.

Sources
Huawei FusionSolarEnphase Enlighten
Data pipeline
Cleaning and normalisationCheckpoint and resumeAPI spend gating
Reports
Annual: yearly KPIsMonthly: seasonal patternsWeekly: trend charts
Upsell agent
BatteryPanel extensionEnergy management system
The pipeline handles the data. The agent applies fixed rules and explains its reasoning.

The two monitoring platforms

Each platform has its own dashboards and API. The pipeline brings their data into one consistent format.

Huawei FusionSolar: energy flow, energy management and module-level monitoring. Vendor screenshots with demo data, from the App Store listing.
Enphase Enlighten: system performance over time. Vendor product image.
Enphase Enlighten: fleet dashboard across systems. Vendor product image.

Key modules

Data pipeline

Collects production, consumption, import and export per installation. Checkpoints let an interrupted run resume, and spend gating keeps API use under control.

Decision rules

Deterministic rules based on thresholds validated by technicians, evaluated in a fixed order. A battery answers a daily mismatch between production and use, and panels answer a yearly shortage.

Recommendations

Each one cites the exact values and sources behind it, so any technician or salesperson can check the reasoning.

Edge cases

Clients close to a threshold are flagged as ones to watch, and installations with too little data are reported as such, with no forced recommendation.

How it was built

The agent is configured with deterministic instructions and runs on each data cycle with the three reports as input. Before applying any rule, it computes intermediate variables such as the number of high-export months and the longest run of months with heavy consumption. These reveal seasonal patterns that yearly totals hide. The documentation reached version 3.2 as the rules were refined with the field team.

Results

  • 80+ installations covered by one pipeline, across two monitoring platforms.
  • Manual, analyst-dependent review replaced by consistent, traceable recommendations.
  • Upsell opportunities are now ranked by urgency, with the evidence attached.