~/iflux/Work

_case-studies~/work

Ideas don't ship.
Systems do.

// every engagement below started as a whiteboard sketch.
Every one of them ended in production.
8
systems shipped
8
industries
0
abandoned in prod
$ filter --tag8 of 8 projects
01Real-time risk engineDecision engine for a trading platform: every order scored against 40+ signals before it hits the book./fintech · /ai40ms → 4msdecision latency
02Enterprise knowledge platformRAG system over 20 years of internal documentation. Answers with citations, or says it doesn't know./ai · /enterprise87%ticket deflection
03Clinical data pipelineTwelve source systems, one HIPAA-compliant warehouse. Auditable lineage on every record./healthtech · /data12 → 1sources unified
04Headless storefront replatformMigrated a mid-market retailer off a legacy monolith with zero downtime during peak season./commerce+31%conversion
05Autonomous regression suiteSelf-maintaining test coverage that rewrites its own selectors when the UI changes./qa · /ai6h → 11mintest cycle
06Warehouse consolidationFourteen ETL pipelines rebuilt as one declarative DAG. Every byte accounted for in the cutover./data0 byteslost in migration
07Subscription billing coreUsage-metered billing for a SaaS scale-up: proration, dunning, and revenue recognition that reconciles./commerce · /fintech99.998%invoice accuracy
08Fraud triage copilotAnalyst-in-the-loop model that ranks alerts by expected loss, not rule count./fintech · /ai-72%false positives

_frequently asked

How many systems has iFlux shipped, and across how many industries?

This index reflects 8 systems shipped across 8 industries over 10 years, from 2016 to 2026. Each entry is a full production build, not a prototype or pilot, spanning categories such as artificial intelligence (AI), data engineering, commerce, and quality assurance (QA) testing. The one constant across all 8: none have been abandoned in production.

How are the case studies on this page selected?

Every entry here is a system that reached production and stayed there — the filterable index groups each one by discipline (AI, data, commerce, QA) and industry (fintech, healthtech, and others) so you can narrow to the closest match to your situation. We don't include work that stalled at prototype stage or was later abandoned; that's the "0 abandoned in prod" stat above.

Why don't the case studies name the client companies?

Client names are withheld by design. Many of our engagements — a risk engine for a trading platform, a clinical data pipeline for a healthcare provider — involve confidential systems under non-disclosure agreement (NDA), so we describe the problem, the build, and the measured outcome (like decision latency or ticket deflection) without naming the company. We can speak to specifics, including references, directly once we understand your project.

Are the outcomes shown here, like faster response times or fewer errors, typical for a new engagement?

Each metric shown, such as cutting decision latency from 40 milliseconds to 4 milliseconds or reaching 99.998% invoice accuracy, reflects the outcome measured for that specific system, not a guaranteed result. Results depend on the starting state of your systems, data, and team. What is consistent across all 8 engagements is that each shipped to production and stayed there, which is the bar we hold every new engagement to.

Can I speak with a past client before starting an engagement?

Yes, for qualified engagements we can arrange a reference conversation with a past client, though some are limited by non-disclosure agreement (NDA) terms tied to the work described above. The best way to start is through the contact page, where we can match you with a relevant reference based on your industry and the type of system you're looking to build.

// yours could be next on this index.

Have a system that needs to exist?

Get in touch →