Forward-deployed engagements

Case studies.

Real work where embedded engineers made inventory visible and ordering obvious, including rare and long-lead parts, without a multi-year ERP program.

Industrial ManufacturingOne quarter to full production

Building an AI-Ready Modern Data Lifecycle

Legacy System Modernization & Workflow Automation

Transformed a $1B solar manufacturer's fragmented inventory and production data into a live view, so planners could see stock and WIP instead of waiting on 24-hour reports.

<30 min

Data latency

99.5%

Pipeline reliability

40 hrs/wk

Time saved

<1%

Error rate

The challenge

Production metrics locked in Oracle MES behind VPN. Financial data scattered across Oracle ERP and SAP. 40+ hours weekly spent on manual reconciliation. 24-hour reporting delays blocking AI initiatives.

Our approach

→Automated VPN access and catalogued 328 Oracle schemas with latency mapping
→Activated change-data-capture (CDC) across Oracle and MES systems
→Built unified event schema preserving order-to-shipment causality
→Containerized self-healing pipeline with automated health monitoring
→Deployed real-time dashboards with LLM-powered translation for supplier docs

Client

U.S. solar technology manufacturer ($1B revenue, 5GW annual production) with strategic partnership with $50B materials multinational.

Systems

Oracle 19c ERPSAP modulesLegacy MES via FortiClient VPNMandarin supplier PDFs

Architecture

Oracle LogMiner CDCAzure PostgreSQL event storeDocker containerized pipelinesGitHub Actions CI/CD

Outcomes

✓Finance reviews live dashboards instead of day-old spreadsheets

✓Planners schedule against real-time WIP visibility

✓Foundation enables demand forecasting and predictive maintenance pilots

✓Zero manual reconciliation required

Energy & Battery Storage16 weeks to production

Unified Operations Platform for Grid-Scale Storage

ERP-MES Integration & Production Intelligence

Connected a battery plant's disconnected systems so production and inventory were visible in real time, cutting order-to-delivery by 35% and stopping planners from waiting on end-of-day stock reports.

Real-time

Order visibility

-35%

Cycle time

100%

Quality traceability

<5 min

Customer updates

The challenge

ERP system (SAP) couldn't communicate with shop floor MES. Production planners relied on end-of-day reports. Quality data trapped in isolated databases. Customers waiting 72+ hours for order status updates.

Our approach

→Built bidirectional integration layer between SAP ERP and proprietary MES
→Implemented real-time production event streaming to unified data warehouse
→Created cell-level quality traceability linking test data to production batches
→Deployed customer-facing portal with live order tracking and delivery ETAs
→Automated compliance reporting for grid-interconnection requirements

Client

Grid-scale battery manufacturer ($500M revenue, 2GWh annual capacity) supplying utility-scale energy storage projects.

Systems

SAP S/4HANA ERPProprietary MESCell testing databasesSCADA systems

Architecture

Kafka event streamingSnowflake data warehouseReal-time API layerAutomated compliance reports

Outcomes

✓Production teams see live WIP across all assembly lines

✓Quality issues traced to specific cell batches within minutes

✓Customers self-serve order status instead of calling support

✓Compliance reports generated automatically for grid certifications

Architectural & Custom Lighting3-week paid pilot

Automating the bill of materials for a custom lighting line

Active design-partner pilot

A European architectural-lighting manufacturer where every order is a unique configuration. Four constructors spend two to three weeks hand-building the BOM for a single order, even though about 80% of the components repeat. The pilot auto-proposes the standard lines so they review instead of rebuild.

2-3 wks

BOM prep today

~80%

of lines are standard

100+

elements per order

3 wks

parts-ordering lag

The challenge

The light planner produces drawings but nothing connects them to production. Constructors rebuild each element in CAD and type the BOM into Excel by hand. About 80% of every BOM is standard components; only 5 to 10 of roughly 50 positions vary. Because the consolidated parts list does not exist until they finish, parts ordering runs about three weeks late.

Our approach

→Ingest the light-planner PDF and DXF plus the Excel constraints sheet
→Mine rules from past approved BOMs to auto-propose the standard 80 to 90%
→Vision model handles the variable positions (housing, diffuser, reflector, LEDs, driver)
→Constructor reviews, edits, and approves every line; the human holds final authority
→Export the approved BOM to Excel in the client's exact column format

Client

European architectural and custom lighting manufacturer. The flagship configurable line is about 40% of production; orders often exceed 100 elements, each its own BOM.

Systems

Light planner (PDF + DXF)Excel constraints sheetLibrary of approved BOMs

Architecture

DXF parsingRules mined from historyBedrock vision modelCelery + S3Tenant-isolated

Outcomes

✓Goal: BOM turnaround from weeks to same-day

✓Constructors redirect to the custom 10% that needs judgment

✓Parts ordering can start before the BOM is fully built

✓Paid from day one: proof of revenue, not a free demo

Discrete / Converter Manufacturing3-week fixed-fee PoC

Attributing every invoice to the right project

Active design-partner PoC

A discrete manufacturer where accepting a single invoice runs a manual loop of about 15 emails and up to 10 people. Poland's KSeF delivers every invoice as structured XML but carries no project or cost center. The PoC reads the XML directly and proposes the project, routing only the ambiguous ones to a single approver.

~15 emails

to clear one invoice today

~10 people

CC'd per invoice

attribution patterns learned

≥60%

auto-verify target

The challenge

KSeF guarantees a tax-validated invoice but nothing about context: no project, no cost center, no order reference. Clearing one invoice today means logging into KSeF, emailing to ask if it matches the order, checking internal records, and only then accepting and assigning a project. About 15 emails and up to 10 people CC'd to clear a single invoice, hundreds of times a month.

Our approach

→Read KSeF invoice XML directly, no OCR: vendor, line items, amounts, and project tokens
→Match each invoice to order records and produce an order-match verdict
→Propose project and cost center with a confidence score using four learned patterns
→Seed the project dictionary from the client's own historical labels
→Route only low-confidence or mismatched invoices to one named approver, replacing the broadcast CC

Client

Discrete and converter manufacturer, part of a multi-company group. System of record enova365; mandatory national e-invoicing via KSeF.

Systems

KSeF invoice XMLOrder recordsenova365 (system of record)Historical reply-email labels

Architecture

XML parser (no OCR)Attribution engine + confidenceReviewer queue + audit trailSynthetic-first; data stays on-site

Outcomes

✓Goal: from 10 people to 1 approver, on under 20% of invoices

✓Order mismatches caught before payment, not after

✓Clean, project-tagged cost data from day one

✓Phase 2 path: true per-project material cost and loss

See what this looks like for your inventory.

Every engagement starts with a discovery call. We will map how you count and order today, especially rare and long-lead parts, and outline a phased approach.

Start discovery Try the demo