// Case Studies

Outcomes,
not adjectives.

A selection of US engagements. Each one was scoped to a measurable outcome and delivered by a senior team in 8 to 16 weeks.

Reduced radiology read time by 38% with FDA-aligned CV triage
Healthcare · Computer Vision
Case / 01
Client · Multi-state hospital network

Reduced radiology read time by 38% with FDA-aligned CV triage

We built a CNN-based triage model on chest X-rays, integrated with the hospital PACS via DICOM, deployed on Azure with HIPAA-aligned VPC and audit trails. Radiologists receive a prioritized worklist; high-acuity findings surface within 60 seconds of upload.

38%
Faster reads
94%
Sensitivity
12wks
To production
StackPyTorchONNX RuntimeAzure MLDICOM/HL7Power BI
$2.1M annual savings via internal research copilot on 4M documents
Financial Services · LLM / RAG
Case / 02
Client · Top-50 US Asset Manager

$2.1M annual savings via internal research copilot on 4M documents

Built a Claude-based research copilot with multi-hop retrieval over proprietary research, filings and call transcripts. Custom eval harness measured faithfulness, citation accuracy and hallucination rate; deployed with SSO, audit logging and per-team access controls.

$2.1M
Annual savings
4M+
Documents indexed
0.91
Faithfulness eval
StackClaude SonnetPGVectorRagas evalsOkta SSOSnowflake
Cut inventory waste 22% with SKU-level demand forecasting
Retail · Forecasting
Case / 03
Client · US grocery chain · 1,400 stores

Cut inventory waste 22% with SKU-level demand forecasting

Hierarchical forecasting at SKU-store-day granularity, blended with replenishment optimization and weather signals. Deployed on Databricks with feature store and online inference; integrated to existing SAP IBP.

22%
Less waste
1.4K
Stores live
8.4%
MAPE
StackDatabricksProphet + LightGBMFeature StoreAirflowSAP IBP
Defect-detection at line speed — 99.2% recall on micro-cracks
Manufacturing · Computer Vision
Case / 04
Client · Tier-1 automotive supplier

Defect-detection at line speed — 99.2% recall on micro-cracks

Custom CV pipeline on edge GPUs (Jetson) detecting sub-millimeter cracks on cast components. Integrated to the existing PLC and MES; flagged parts route to manual QA in real time.

99.2%
Recall
120ms
Per-part inference
6wks
POC to pilot
StackPyTorchTensorRTJetson OrinOPC-UAGrafana
Shipped in-product copilot, lifted activation 31%
SaaS · Generative AI
Case / 05
Client · B2B SaaS · Series C

Shipped in-product copilot, lifted activation 31%

Co-built an in-product GenAI assistant with tool-use over the SaaS app's APIs. Includes guardrails, prompt-injection defenses, and a robust eval gate before each release.

31%
Activation lift
+18 NPS
Trial users
9wks
Idea to GA
StackGPT-5.4Vercel AI SDKPostHogPromptfooDatadog

Want similar outcomes?

Made with Emergent