Book a meeting

15 case studies

See the work behind the expertise.

Explore engineering work across semiconductor manufacturing, property data, title insurance, media and energy. Find a relevant problem, then review UTIS’s contribution and the result.

Filter by industry, challenge or platform. Each case retains its delivery stage and measurement scope.

Industry

15 of 15 case studies

Refine by challenge or platform

Problem type

Platform focus

Explore the Databricks practice

Conceptual summary

BeforeFull-table reloads
AfterIncremental PySpark jobs

Semiconductor

Cut Databricks compute cost by ~57% across a fleet of machine-log pipelines

Standardized deployment with Databricks Asset Bundles and a Terraform-based pipeline, so every pipeline ships the same way.

~57%

lower average daily compute cost across the eleven pipelines (first days after the final migration vs. a five-month baseline).

How we measured it: platform-reported daily compute cost, averaged over the baseline period and the initial post-migration window; the same datasets continued to reach users throughout.

Conceptual summary

ContextGeographic scope
UTIS contributionSource-data checks

Production checks confirmed

Property data

Property reporting that stands up to a source-data check

Client-validated

Reporting changes checked against source data after production release.

Evidence basis: client production checks and completed work records; full scope in the case.

Conceptual summary

ContextData definitions
UTIS contributionCortex Analyst application

Delivery confirmed by client

Media

Business questions, answered through a Snowflake application

Client-confirmed

Snowflake integration and Cortex AI development confirmed by the client.

Evidence basis: client confirmation, delivery and QA records; full scope in the case.

Conceptual summary

ContextSQL + ETL data
UTIS contributionPower BI reports

Report publication confirmed

Title insurance

Power BI reporting leaders can trace back to the data

Published reports

Power BI delivery confirmed by the client, with source-data checks for selected measures.

Evidence basis: client-confirmed publication and SQL checks; full scope in the case.

Conceptual summary

ContextFile deliveries
UTIS contributionData-load checks

Documented delivery

Media

Keeping media analytics moving as source data changes

Production delivery

Warehouse updates and a client-confirmed S3 monitoring component.

Evidence basis: production work records and client confirmation; full scope in the case.

Conceptual summary

ContextAccess requirements
UTIS contributionPermissions design

Implementation scope

Energy

A customer portal with architecture ownership from the start

Implementation

Power Pages portal progressed beyond proof of concept with a UTIS technical architect.

Engagement summary: architecture and implementation work; full scope in the case.

Conceptual summary

ContextExisting stream
UTIS contributionShared health monitoring

Existing platform support

Semiconductor

Monitoring and access for an existing machine-data stream

UTIS operates and extends an existing streaming path, with consolidated health monitoring, machine onboarding and self-service access.

100%

of the existing streaming path’s components visible in one health dashboard.

Existing-platform monitoring, onboarding and access work; the latency is operating context.

Conceptual summary

ContextManual re-creation
UTIS contributionLLM/NLP conversion

Production rollout in progress

Semiconductor

~70% less manual work preserving engineering knowledge from a retiring system

Co-built, with the client team, an LLM/NLP pipeline that reads the legacy XML and generates the decision logic and configuration rules the new platform needs.

~70%

less manual workload (joint client–UTIS team result).

How we measured it: accuracy assessed by the joint team against expert-authored references; the workload and review-time figures are the team's estimates.

Conceptual summary

ContextShared workstations
UTIS contributionDatabricks App and Azure AKS

Workflow cutover in progress

Semiconductor

Browser-based factory analysis validated against existing outputs

Browser job submission, monitoring and output review were validated against the existing system; daily-workflow cutover remained in progress.

Outputs matching

the existing system. Daily-workflow cutover in progress at time of writing.

How we measured it: output reconciliation against the existing production system.

Conceptual summary

BeforeStandard environment
AfterAzure AKS GPU pipeline

Validated proof of concept

Semiconductor

A scientific AI workload proven ~10× faster on a locked-down GPU platform

Built the first end-to-end machine-learned simulation pipeline on the platform (NVIDIA ALCHEMI and NIM, PyTorch, MACE), reproduced the known physics on H100 GPUs and benchmarked it against the standard environment.

~10×

faster than the standard environment on the same physics, in a validated proof of concept.

How we measured it: benchmark against the standard environment with the same inputs; the reference physics was reproduced before timing was compared.

Conceptual summary

BeforeSSIS processing
AfterSSIS and daily absence checks

Incident-proven detection

Semiconductor

Next-day visibility into missed data processing

UTIS improved SSIS processing and added a check for the previous day’s unprocessed files; a real incident confirmed next-day detection.

Next day

Missed processing detected in one documented incident.

Incident-proven next-day detection; response and service commitments are scoped separately.

Conceptual summary

BeforeHand-set thresholds
AfterValidated statistical limits

Semiconductor

Factory alerts grounded in sensor history, and an out-of-tolerance machine caught before scrap

Our business analyst ran structured sessions with operators and area managers to define the thresholds the business could stand behind.

one incident

During development, an out-of-tolerance machine was stopped for inspection before it produced scrap.

How we measured it: one observed incident; the savings are projections from the agreed business case.

Conceptual summary

BeforeMajor-version upgrade
AfterPython comparison

Semiconductor

A data-platform upgrade validated across 2,000+ views before anything reached production

Designed and led a Python regression-automation framework: automated old-versus-new output comparison, row-level and aggregate reconciliation, discrepancy detection and root cause.

2,000+ views

validated automatically; migration defects caught before production; go-live green.

How we measured it: automated comparison results at go-live.

The illustrations are conceptual. Results describe the individual engagements; client identities remain confidential and references require client permission. The Databricks filter includes cases whose main delivery scope uses the platform; separate MLOps work is described on the practice page.

Meet with UTIS

See a challenge that feels familiar?

Tell us what is happening in your environment. We’ll discuss the relevant experience, the constraints and a practical way to begin.

Choose a preferred time

Book a meeting

UTIS will review your preferred time and confirm the meeting by email.