Lumera Energy

Inspiration

Why are home energy systems still controlled by fixed rules when home conditions change all the time?

A modern home can have solar panels, a battery, AC, EV charging, changing electricity prices, and changing weather. But most home energy systems are still rigid. They follow simple automations, react poorly when conditions shift, and force the homeowner to keep stepping in manually.

We wanted to build something better: a system that does not just automate, but also adapts. That idea became Lumera Energy.

What it does

Lumera Energy is an adaptive AI home energy agent.

It monitors:

time of day
grid electricity price
outdoor temperature
synced external context

On every cycle, Lumera selects exactly one action, such as:

storing energy in the battery
discharging the battery
buying from the grid
selling back to the grid
charging the EV
pausing EV charging
shutting off AC
restoring AC

If the homeowner disagrees with a decision, they can override it. Lumera treats that override as feedback, converts it into a persistent preference guideline, and uses that learning in future decisions. That makes the system more adaptive and more personal over time.

How we built it

We built Lumera as a real-time agent loop using TypeScript, Node.js, Express, and PostgreSQL.

The system simulates live streams such as:

Clock
Grid_Price
Weather_Temperature

On each update, the agent:

reads the latest stream values
reads external context from PostgreSQL
chooses exactly one action
applies active preference guidelines
sends the action and the main reason to the dashboard

We used Airbyte as an external data connector. It syncs outside context into PostgreSQL so the agent can use more than just local sensor values when making decisions.

We used Auth0 to secure dashboard access and protect sensitive user actions like overrides and preference changes.

We used Ghost as the publishing layer for summaries and reports, so the homeowner can clearly see what the agent decided and what it learned.

The frontend is a live dashboard built with HTML, CSS, JavaScript, and Server-Sent Events (SSE).

Challenges we ran into

One challenge was making the system feel like a real agent instead of a simple rules engine. We wanted Lumera to choose exactly one action every cycle, apply learned preferences, and still stay stable when data changed quickly.

Another challenge was integrating the sponsor tools in a natural way. We did not want Airbyte, Auth0, and Ghost to feel forced or disconnected from the product. Each had to play a clear role:

Airbyte for external context
Auth0 for secure control
Ghost for transparency and reporting

We also had to think carefully about trust. A home energy agent cannot just make decisions. It also has to explain them clearly and keep the human in control.

Accomplishments that we're proud of

We are proud that Lumera is more than a static smart-home demo.

It:

makes real-time energy decisions
learns from homeowner overrides
uses Airbyte-synced context from PostgreSQL
secures user control with Auth0
publishes transparent summaries through Ghost

Most importantly, it has a real agent loop: observe, decide, explain, accept feedback, and improve.

What we learned

We learned that good agent systems need more than decision logic. They also need context, trust, and feedback.

Airbyte showed us how useful synced external context is for better real-time decisions. Auth0 showed us that secure control is a core product feature, not something to add later. Ghost helped us think more seriously about transparency, because people trust systems more when they can clearly see what the system did and why.

We also learned that user overrides are powerful. Instead of treating them as interruptions, we turned them into a learning signal. That made the product feel much more adaptive and personal.