Our work

Systems shipped in the operation, not in a demo.

Selected projects led by Pala's founder in previous operating roles. Metrics are from those roles. No product screenshots are shown — none are cleared for public use, and this page does not use placeholder images that look like product UI.

Featured · AI automation

AI-powered accounts payable

LLM-powered invoice automation built in an operating role at RepairWise. The ledger stayed in place; the work was intake, rules, routing, and the path to payment.

~$150K estimated annual savings

Problem

Invoice intake, coding, and approval routing were eating recurring finance time. Exceptions stalled in email instead of a controlled workflow.

What was built

An LLM-powered accounts-payable workflow that processes invoices, applies business rules, and automates the repetitive finance steps around the existing ledger.

How it works

  1. Invoices are extracted from the documents finance already received
  2. Business rules and duplicate checks run before anything posts
  3. Approvals route from those rules instead of an inbox
  4. The handoff goes into QuickBooks
  5. Payout anomalies are flagged before money leaves

Tech and approach

LLMs for document processing, then deterministic rules for duplicates, approvals, and payout checks. This is not a new ERP, and it is not the OEM invoice-matching app described elsewhere. The published stack beyond “LLM-powered” is the workflow above; the specific model is not named in public sources.

LLMs Document processing Business rules Finance automation

Company not named. Throughput, exception rate, and model vendor are not in the public sources used here. Existing AP write-up →

Custom software

Partner operations portal

Problem

External partners and internal operations were coordinating through fragmented workflows and manual follow-up.

What shipped

A custom platform connecting partner shops with internal operations. Launched across 15 shops. Related shop-network tooling later included inventory attestation, parts requests, and RMA paths.

400+jobs supported

Custom applications Workflow automation Integrations

AI agents

AI labor-coding engine

Problem

Repair labor coding was slow, repetitive, and sat on the path to payout.

What shipped

A labor-code prediction API with an automated weekly feedback loop against actuals, so the system keeps learning from what actually got coded. Built for technicians and operations, not as a one-off script.

20%faster coding and payouts

AI agents Business rules Workflow execution

Limited public detail

Custom software

Internal CRM

Live internal CRM for contacts, locations, and accounts, with Google Places enrichment. Next.js and Supabase. Not a full case study — usage and what it replaced are not in the public sources.

Next.js Supabase Places API

Also shipped

Other production systems from the same operating stretch. Status as labeled on the public work log.

OEM invoice matching

A separate finance system: AI-assisted matching of OEM invoice line items to orders. The model proposes matches and flags; finance confirms before anything is escalated. This is not the AP automation case study above, and it is not the $200K+ excess-inventory figure.

Shop inventory loop

Attestation, keep/return recommendations, RMA, and 518 structured parts requests with defect flags. Confirmed result: $200K+ excess inventory identified.

Network assignment

Distance-based shop assignment over Census ZCTAs. 89% historical assignment agreement in backtesting; not a live accuracy claim.

Shop-message copilot

Grounded reply drafts and workflow recommendations on CRM shop messages. Beta.

About 30 internal tools were built at RepairWise / Fixlane in the last year, mostly ops tooling. Claude shows up in extraction, keep/return recommendations, and the copilot stack. No screenshots from those systems are published here.

Want something in this shape for your operation?

Start with the workflow, not a tool category.

Discuss a Project