Energy & Utilities practice

Applied AI for the regulated utility.

We harness frontier models to build agentic workflows for the utility, delivered by forward-deployed engineers on a context and trust layer we bring to every engagement.

Expertise Customer operations Grid & outage Field & assets Corporate

Start a conversation See the platform

AWS Bedrock

GCP Gemini / Vertex AI

Domains

One practice across the utility, starting where value is proven fastest.

Customer operations

Customer service agent, high-bill alerts, collections and assistance matching, outage communications. All on Customer 360 and shared MCP tools.

In production

Grid & outage operations

Dispatch copilot, outage ETR prediction, relay-settings review, wildfire risk, and a grid digital twin for what-if and switching studies.

Shared foundation

Field, assets & programs

Field safety briefings, transformer and asset health, DER interconnection queues. Program targeting for DR, efficiency, and rate plans from AMI data.

Roadmap

Corporate

Legal review and contract management with the clause cited, regulatory filings, and analytics agents that answer questions over the lakehouse in plain language.

Shared foundation

The fleet

Agents at every voltage level, from the plant to the meter to the back office.

Generation & transmission

Grid opsOperator copilot over SCADA, EMS, and switching orders.
Relay settingsProtection setting review and coordination checks against the network model.
Wildfire riskCorridor risk from weather, vegetation, and asset condition; PSPS scoping.
Digital twinNetwork model for what-if, switching, and hosting-capacity studies.

Operations & field

DispatchCrew assignment and storm-mode triage from OMS and crew notes.
Outage ETRPremise-level restoration estimates that beat the crew guess.
Field safetyJob briefings, hazard history, and clearance checks per work order.
Asset healthTransformer and feeder condition from load, thermal, and inspection data.

Customer

CS agentDeflection on top call drivers, CSR copilot, 100% QA.
High-bill alertPredict and explain the bill from interval data and tariff.
CollectionsPropensity-to-pay, assistance matching, compliant arrangements.
DER interconnectInterconnection applications triaged, hosting capacity answered.

Corporate

Legal reviewContract and filing review against policy, with the clause cited.
Contract managementObligations, renewals, and rate terms tracked from the signed PDF.
AI for BIPlain-language questions over the lakehouse, with the SQL and lineage shown.
Regulatory filingsRate-case and PUC data requests drafted from certified sources.

Every agent is built for the utility it runs in. What stays constant is the context layer it reads from and the certification evidence it ships with.

What we bring

Four pieces we bring to every engagement.

Customer 360

Resolved CIS, MDM/AMI, OMS, payments, and contact center, governed and lineaged.

Unity Catalog

Customer Service AI Agent

Deflection of top call drivers, a CSR copilot, and QA scoring on 100% of interactions.

Voice · chat · digital

Five value streams

Service, meter & billing, outages, collections, and contracts, all reading the same context.

One tool layer, many agents

Agent Certification

Every agent cataloged, scored, human-calibrated, and certified before it talks to a customer.

Audit-ready evidence

The platform

Build the context once. Every agent reads from it.

Trust

Agent certification. Every agent scored against a versioned domain rubric, re-certified on change. The certificate is what goes to the regulator.

Certification · Interaction QA

Agents

Many agents, one integration. Customer, operations, and corporate agents call the same MCP tools over CIS, AMI, OMS, SCADA, work management, and the contract library instead of integrating each system again.

Databricks Apps · Model Serving

Context

Agent Context Platform serves pre-joined entity profiles over MCP. Digital Brain retrieves transcripts, notes, tariffs, and contracts with citations. Numbers come from tools, never from retrieved text.

ACP (Lakebase) · Vector Search

Foundation

Customer 360 and asset gold layers plus a feature store: balance, 13-month usage, 90 days of interactions, outage exposure, asset condition, risk scores.

Unity Catalog Delta

Customer operations · five value streams

Where the money and the complaints already are.

Customer service

Intent analytics on every transcript, a real-time CSR copilot, and QA on 100% of interactions.

Furthest along

Meter & billing

Predict the high bill before it lands. Explain it from interval data and tariff, not a guess.

Risk · medium

Outages

Premise-level ETR, storm-mode self-service, medical-needs customers notified first.

Risk · low

Collections

Propensity-to-pay, assistance matching, compliant arrangements. A human approves.

Risk · high

Contracts

Obligations, renewals, and rate terms extracted from PDFs with the clause cited.

Risk · low to high

North star

Every initiative hangs on an outcome leadership already tracks.

Cost-to-serve

Calls per customer, AHT, containment.

CSAT / J.D. Power

Proactive, personalized bills and outage comms.

Bad debt & DSO

Early intervention instead of blanket dunning.

PUC compliance

100% interaction scoring, certified agents.

Trust in AI

Agent certification as standing evidence.

Sequencing

Foundation first. Customer-facing agents only after certification evidence exists.

H1 · months 0–6

Foundation and the biggest call driver

Customer 360 gold layer + feature store
Intent analytics on all transcripts
CS AI Agent certified before launch
High-bill alert model
Digital Brain v1 + ACP

Gate: C360 live, CS agent certified

H2 · months 6–15

Copilot and prediction, human in the loop

CSR copilot on Customer 360 MCP tools
Outage ETR prediction + personalized comms
Propensity-to-pay + assistance matching
Dispatch copilot + field safety briefings
AI for BI over the lakehouse
Human-agent QA scoring

Gate: copilot labels calibrate the judges

H3 · months 15–24

Customer-facing, transactional, governed

Customer 360 agent on digital and voice
Self-service transactions: payment plans, enrollment
Asset health, relay review, grid digital twin
Legal review + contract management
Fleet governance across all agents

Delivery

Forward-deployed engineering, not a requirements document.

Agentic workflows fail in the gap between the spec and the floor. So we close the gap: our engineers sit with the people who will use the agent, build on your governed data from the first week, and certify as they go.

Traditional build

Forward-deployed

Where the team sits

A build team behind a project manager and a requirements document, often offshore.

Engineers embedded with dispatch, billing, and the contact center, on-site or in their daily standups.

First working software

After discovery, design, and sign-off. Usually a quarter or more.

A working agent on your governed data inside the first two weeks. Scope is set by what the users do with it.

Feedback loop

Quarterly demos and a change-request queue.

Daily, with the CSRs, operators, and analysts who use the agent. Every shift is a test run.

Data

Sample extracts and synthetic records until go-live.

Your lakehouse from day one. Context is built once and every agent reads from it.

Evaluation

User acceptance testing at the end of the project.

A certification rubric written in week one and scored continuously. Certification evidence is a deliverable, not a phase.

Hand-off

Documentation and a support contract.

Your engineers pair with ours and ship the next agent on the same context layer without us.

Commercials

Time and materials against a spec that drifts.

Fixed-scope sprints tied to one outcome the business already tracks.

Weeks 0–2

Embed and baseline

Shadow the floor, pick one call driver or one outage pattern, stand up the context layer on your data, write the rubric.

Weeks 3–8

Build with the users

Ship a working agent to a pilot group, iterate daily on real interactions, score every change against the rubric.

Weeks 9–12

Certify and hand over

Produce the certification evidence, go live under a human-approval gate, and pair your team onto the next agent.

“Real artists ship.” Steve Jobs, to the Macintosh team, 1983

Partner

Built on Databricks, end to end.

Technology partner

Databricks is the data, ML, and agent plane for the practice. Unity Catalog governs Customer 360, Vector Search powers the Digital Brain, and the agents, their context, and their serving all stay inside the workspace.

Lakebase

The Agent Context Platform runs on Lakebase. Entity profiles live as Postgres JSONB with pgvector, rebuilt from the gold layer by schedule and change feed.

Postgres · pgvector · changefeed

Agent Bricks

Where the agents are built and tuned. Digital Brain retrieval agents and the CSR copilot are composed, evaluated against the certification rubric, and iterated here.

Model Serving · MLflow evals

Databricks Apps

The MCP tool layer and the copilot UI ship as Apps, so explain_bill and get_outage_status run next to the data with Unity Catalog permissions intact.

MCP tools · CSR copilot

Deployment

Runs in your cloud, on the model platform your security team already approved.

AWS

Amazon Bedrock

Agents served through Bedrock with Guardrails, inside the utility's VPC. Landing zones, IAM, and PrivateLink to the lakehouse.

Microsoft Azure

Azure AI Foundry

Foundry agents and model catalog, Entra ID for every tool call, and a direct path from Dynamics and Teams into the CSR copilot.

Google Cloud

Gemini on Vertex AI

Gemini models on Vertex AI for long-context contract and transcript work, with Cloud SQL and BigQuery federated into Unity Catalog.

Go from one agent to ten without a ten-month review each time.

We work with utility operations, data, and compliance teams on Databricks across AWS, Azure, and Google Cloud. Bring a call driver, a bad-debt number, or a PUC question.

Talk to the practice practice@kartha.energy