Freight Audit & Pay: Legacy to Production in 6 Months
How AI-augmented engineers rebuilt the invoicing backbone of a U.S. delivery orchestration platform, and kept building.
How AI-augmented engineers rebuilt the invoicing backbone of a U.S. delivery orchestration platform.
- Logistics
- FinTech
- Brownfield
- Industries
- Logistics, FinTech
- Platform
- Web application (SaaS)
- Delivery scope
- AI-accelerated engineering, legacy replacement, financial logic, brownfield integration
ABOUT THE PROJECT
Project overview
- Engagement
- Ongoing (1 year+)
- Team size
- 6, joint with the client team
- Core technologies
- .NET/C#, Kubernetes
A U.S. delivery orchestration platform managing enterprise last-mile fulfillment. Billing complexity across shippers and logistics providers outgrew the legacy system — the team needed a purpose-built Freight Audit & Pay platform for multi-party invoicing, cost calculations, and dispute management at scale.
Limestone's AI-augmented velocity pod took full ownership of delivery, shipping a production-grade replacement integrated into the client's existing infrastructure.
A U.S.-based delivery orchestration platform helping enterprises manage last-mile fulfillment with a connected courier network and real-time visibility.
Billing complexity grew across shippers and logistics providers. The team needed a purpose-built Freight Audit & Pay platform.
Limestone embedded a velocity pod of AI-augmented engineers who took full ownership of delivery.
Challenge
Financial-grade logic at scale, without breaking what works. A new invoicing system had to ship fast, stay auditable, and never break the workflows already running underneath it.
-
Multi-party billing at scale
Replace the legacy Knack-based invoicing system with a scalable platform supporting billing across shippers and logistics providers.
-
Financial-grade consistency
Complex cost-calculation logic with tight data consistency requirements across multiple systems.
-
Brownfield constraints
Existing workflows and downstream consumers could not break during migration. Legacy behaviors were mapped and preserved before the new system took over.
Existing workflows and downstream consumers could not break during migration. Legacy behaviors were mapped and preserved.
-
Speed with auditability
Requirements evolved as discovery surfaced edge cases, under pressure to ship value fast while maintaining auditability, security, and operational stability.
Requirements evolved as discovery surfaced edge cases, under pressure to ship fast while maintaining auditability and stability.
WHAT WE DID
Solution
The Velocity Framework: six steps, every feature, every time.
-
Invoice Generation
Traditional invoices for shippers and reverse invoices for logistics providers, integrating directly with delivery data.
-
Cost Calculation Engine
Complex cost-targeting logic with stored properties, automatic recalculation on delivery changes, and support for Traditional and PostAdjustment invoice types.
Complex cost-targeting logic with stored properties, automatic recalculation on delivery changes.
-
Dispute Management
Full dispute workflow with status tracking, partial approvals, and resolution flows built from real auditor feedback.
-
Approval Workflows
Configurable approval windows with Kubernetes CronJobs that automatically process expired invoices daily.
-
Reporting & Export
Detailed invoice reporting with configurable columns, CSV export, and full transparency into dispute values.
-
Event-Driven Architecture
Event-driven state changes keep downstream systems synchronized with billing activity in real time.
-
Cursor
Primary development environment with Plan Mode.
-
Claude Opus
Primary model for spec writing, plan generation, and code implementation.
Primary model for spec writing and code implementation.
-
ClickUp
Task management and requirement tracking, the starting point for every feature.
Task management and requirement tracking.
A reliable, event-driven backbone.
Event-driven microservices on Kubernetes with message-based synchronization.
Relational persistence and integrations across the delivery and data infrastructure. Event-driven state changes keep downstream systems synchronized.
Relational persistence and integrations across the delivery and data infrastructure.
DELIVERY TIMELINE
From zero to production, then kept building.
Upper environments at roughly four months, production by months five to six, and the same Limestone team continues expanding the platform.
Upper environments at roughly four months, production by months five to six.
-
WEEK 1-2
Context & Onboard
Knowledge graph first
-
MONTHS 1-3
Core Build
122 PRs in 90 days
-
MONTH 4
Upper Environments
Ahead of plan
-
MONTHS 5-6
Validation & Prod
330+ PRs merged by month 6
330+ PRs merged
-
MONTH 7+
Ongoing Development
Same team, still shipping
DELIVERY APPROACH
In production ahead of plan. Brownfield, not a weekend prototype.
-
WEEK 1
Context before code
No code in the first week. The team scanned the entire repo and built a knowledge graph documenting every module, dependency, and data flow.
-
V.U.E.
The V.U.E. gate
The agent handles roughly 90% of code generation. A senior engineer reviews every PR, nothing merges unless they can verify, explain, and debug it.
The agent handles roughly 90% of code generation. A senior engineer reviews every PR.
-
50% FASTER
Where the time savings came from
Upfront context eliminated false starts, the agent handled mechanical coding so engineers focused on design and review. AI cost: $200/month
CLIENT FEEDBACK
The client paid the engineers discretionary bonuses. Twice.
-
TECH LEAD
"True engineers capable of closing requirement gaps through fast domain discovery and iteration."
"True engineers capable of closing requirement gaps."
Tech Lead · Delivery Orchestration Platform · Anonymized
Tech Lead · Delivery Orchestration Platform
-
PRODUCT TEAM
"Product maturity level achieved unusually early."
Product Team · Delivery Orchestration Platform · Anonymized
Product Team · Delivery Orchestration Platform
-
VELOCITY STATS
122 merged PRs in the first 90 days
104 merged with fewer than 5 reviewer comments. 90% code generated with AI, reviewed by humans.
104 merged with fewer than 5 reviewer comments. 90% code generated with AI.
ENGINEERS RECEIVED DISCRETIONARY BONUSES FROM THE CLIENT TWICE FOR EXCEPTIONAL DELIVERY
Impact & Results
- Freight Audit & Pay platform deployed to upper environments in four months, in production by months five to six, still actively developed a year later.
- Freight Audit & Pay platform deployed to upper environments in four months, in production by months five to six.
- Legacy invoicing system replaced with a scalable, integrated platform.
- 122 pull requests merged in the first 90 days with consistent quality; 104 required fewer than 5 reviewer comments.
- 122 pull requests merged in the first 90 days; 104 required fewer than 5 reviewer comments.
- 2 AI-augmented engineers maintained consistent high-velocity delivery.
- Engineers received discretionary bonuses from the client twice.
- Velocity Framework validated as a repeatable delivery system on a complex brownfield engagement.
- Velocity Framework validated as a repeatable delivery system.
-
PRs merged in first 90 days
122
-
AI-generated code, human-reviewed
90%
-
faster development cycles
50%