Case studies

Real work, told plainly.

We don't name customers. Their operations and their data are their business. Everything below is a real engagement described accurately, including where it currently stands. No inflated outcomes, no wall of logos. If a case study says "in proposal," that means they haven't signed yet, which is apparently a radical level of honesty for an AI company's website.

In production

Quoting intelligence for a Gulf Coast steel fabricator

The situation

A structural steel fabricator quoting dozens of jobs a month, with estimating knowledge concentrated in a handful of senior people and job history locked inside QuickBase. Every quote was rebuilt from scratch. Margin outcomes were only understood after jobs closed, if at all.

What we built

A quoting and margin platform that syncs continuously from their QuickBase system of record, finds genuinely comparable historical jobs (matched on structural profile and labor density, not keywords), and puts material nesting optimization inline in the estimate builder. Every recommendation is backtested against jobs with known outcomes.

Why it's different

Estimators get a comparable set they can defend in a quote review. The platform reads with minimum-privilege access, never writes to the system of record, and runs in a fenced-off environment with database-level isolation.

Where it stands

Deployed and in production use. The pattern of sync, similarity, and margin analysis is the template for our quoting agents across fabrication and manufacturing.

In development

Policy-to-regulation mapping for a compliance software provider

The situation

A compliance software company whose customers upload internal procedures, manuals, and policies, thousands of pages, that need to be tracked against PHMSA pipeline regulations. Mapping them manually is slow, inconsistent, and doesn't scale.

What we're building

An AI pipeline that ingests customer documents, generates a structured table of contents, and maps each section to the governing PHMSA regulations. Results are delivered back into the provider's own platform by API, without Junipix ever having direct access to their database.

Why it's different

Where the AI can't confidently link a section to a regulation, it doesn't guess. It flags the item into a human review queue. Accuracy over theater. This is our regulatory library working as a component inside someone else's product.

Where it stands

Under contract and in development. The same document-to-regulation mapping powers our policy gap analysis for operating companies.

Pilot proposed

FERC tariff intelligence for a midstream pipeline operator

The situation

A midstream operator whose commercial and regulatory teams work daily with FERC tariff filings. Dense, technical documents where the difference between two tariff provisions is real money, and finding the controlling language takes hours of expert time.

What we proposed

A domain-constrained tariff intelligence pilot: FERC filings and tariff records made searchable and answerable in plain English, with the exact provision cited, scoped to the operator's own systems and workflows.

Why it's different

General-purpose AI paraphrases tariffs from memory. This system answers only from the authoritative filings, cites the provision, and says so when the answer isn't in the documents.

Where it stands

Pilot proposal delivered; in active conversation.

In proposal

Labor relations intelligence for a major union employer

The situation

A major union employer whose labor relations teams navigate collective bargaining agreements, company policies, and past practice across multiple workgroups. Answering "what does the contract require here?" means senior specialists searching hundreds of pages while the clock runs on grievances and operational decisions.

What we proposed

A domain-constrained intelligence system over the agreements and policies themselves: plain-English answers with the exact article and section cited, applicability explained by workgroup, and the strictest-answer logic applied where contract and policy overlap.

Why it's different

In labor relations, an uncited answer is unusable. Every response traces to contract language. It's the same approach we use for government regulations, applied to negotiated agreements.

Where it stands

In proposal. Enterprise engagements at this scale run on longer cycles, and we'd rather say that than imply otherwise.

In proposal

Traceability, safety, and logistics agents for a national food distributor

The situation

A national distributor facing FDA FSMA 204 traceability requirements across thousands of SKUs, suppliers, and facilities. Add the everyday grind of freight negotiations, accessorial fees, and warehouse safety questions that each pull time from experts.

What we proposed

A family of agents on shared infrastructure: FSMA 204 applicability mapping at the SKU level, supplier readiness scorecards, an OSHA-plus-company-policy strictest-answer agent for warehouse supervisors, and logistics negotiation agents that benchmark lanes and flag fee leakage.

Why it's different

One regulatory backbone serving many workflows. It's the same pattern that lets an answer turn into an action: a notification drafted, a supplier flagged, an exception queued for human review with a full audit trail.

Where it stands

Proposed and in conversation. This engagement is where several catalog agents were first defined.

Your company could be the next one we can't name.

Every engagement above started the same way: a conversation about a specific operational problem, then a $2,500 audit that put numbers on it. The conversation part is free.