ClaimGuard
A platform that helps freight carriers recover money they are owed — reading their paperwork, building the claim, and tracking it to payment.
- My role
- Sole developer — product, platform, AI subsystems and security.
- Built for
- Freight & logistics
- Project type
- Claims recovery platform
- Built with
- Next.js · Supabase · Stripe · Anthropic Claude · Vector search · Supabase Storage · Redis rate limiting
The story
What problem existed, and what we built.
The problem
Carriers are routinely owed money for detention, layovers and short pay. Recovering it means finding the right documents, reading the fine print, assembling a claim, and chasing it. Most small carriers simply don't, because the effort per claim is larger than the claim feels.
What I built
A platform that takes the paperwork, extracts what matters from it, walks the carrier through building the right kind of claim, and then manages that claim through to payment.
What it had to achieve
- Make a claim worth filing by making filing it cheap
- Read the paperwork so the carrier doesn't have to
- Keep one clear path into the product
- Treat other people's shipping documents as seriously as they deserve
The customer's experience
A carrier uploads the paperwork for a load. The system pulls out the details that matter, guides them through the right claim, and then tracks it. What used to be an afternoon of document archaeology becomes a short guided flow.
What customers can do
- A guided wizard covering every claim type, with one front door instead of several
- Upload documents and have the relevant details pulled out automatically
- Track claims through to payment
The owner & staff experience
Claims, documents and their status are all in one platform, with payments and notifications wired through it.
What the business can do
- Document intelligence that reads uploaded paperwork
- A retrieval system that can answer questions against a claim's own documents
- An audit trail designed so sensitive question text stays out of the permanent record
- An admin console with two-factor sign-in for review, payments and staff-assisted intake
- Payments, email, rate limiting and bot detection wired in
How it works
Architecture & integrations.
How the application, data, access controls, and connected services work together.
Documents in, structured facts out
Uploaded paperwork is classified and read into the specific fields a claim needs: rules first, a language model where rules can't decide. Staff confirm document types and approve or correct extracted fields before they count, and AI-drafted letters are only sent by a person.
Answers grounded in the actual documents
The document-retrieval subsystem combines vector search with claim-level scope checks. Retrieved excerpts are checked against the requested claim before they can be used as answer context.
One front door
Several older self-service entry points were deleted and now redirect to a single wizard. Multiple half-maintained intake paths are how products quietly break. Staff-led Assisted Intake is a deliberately separate path for cases a team member files on a client's behalf.
Design decisions
The choices that shaped it — and why.
Rebuilt intake on one engine
Every self-service claim type now runs through the same schema-driven wizard. The dead legacy forms were deleted rather than left in place, so there is one thing to fix when something is wrong.
Releases are explicit
Merging to the main branch produces a preview here, not a release. Shipping is a deliberate, separate act — which prevents work-in-progress reaching customers by accident.
Privacy in the audit trail
The permanent audit record deliberately excludes the text of administrator questions. An audit log is supposed to prove what happened, not become a second copy of sensitive content.
How it was built
Development process
- 1Built the two-service platform and the freight lead engine
- 2Added document intelligence
- 3Built and shipped the document-retrieval subsystem, with evaluation gates it has to pass
- 4Hardened retrieval so document scope is verified before any excerpt is returned
- 5Rebuilt self-service intake so all claim types run through one wizard
- 6Implemented retrieval evaluation gates for citation relevance
Technical overview
Technology & implementation
Technology stack
- Next.js
- Supabase
- Stripe
- Anthropic Claude
- Vector search
- Supabase Storage
- Redis rate limiting
Implementation scope
Sole developer — product, platform, AI subsystems and security.
Related projects
You might also want to see
J Kiss LLC
Operations platform for a service business · Project
jkissllc.comA Dallas–Fort Worth hauling company that takes quotes, books jobs, dispatches crews, hires staff and runs payroll from one connected system rather than six disconnected ones.
Read the case study
Want something like this for your business?
Tell me how your days go and what you'd like to stop doing by hand. I'll tell you honestly what would help.