How HomeFront Group 10x'd Their Veteran Caseload Without Adding Staff

Every veteran disability claim at HomeFront Group took days of manual document review. Attorneys read through hundreds of pages of medical records, cross-referenced VA regulations, and prepared legal briefs by hand. The work mattered, and nearly all of it was repetitive.

That left the firm with a lopsided split: attorneys were spending 80% of their time on document analysis and only 20% on actual legal strategy. Veterans waited months for representation they urgently needed.

The scale of the problem

• Client data was scattered across HubSpot CRM, medical record systems, and VA portals, with no single source of truth

• Each case required manually applying the Caluza Triangle framework (diagnosis, in-service event, nexus) across dozens of conditions

• Cross-referencing against 38 C.F.R. regulations and the M21-1 manual took hours per condition

• Evidence gaps were discovered late in the process, causing costly delays and re-filings

• Hiring more paralegals was not economically viable for the firm

We embedded a 5-person team in HomeFront Group's workflow for 7 months, building in increments and validating every capability against real cases.

Phase 1: Mapping the workflow

Before writing a line of code, we sat with attorneys and mapped their claims process end to end. We were looking for the specific bottlenecks where AI could take the most work off them while leaving legal judgment with the attorney.

Phase 2: Data infrastructure

We built the integration layer first, connecting HubSpot, medical record systems, and VA sandbox APIs into a normalized data model. Conditions are organized by organ system, diagnostic codes link to CFR citations, and imports are tracked with version control. Everything built later reads from that model.

Phase 3: AI analysis engine

The core engine uses prompt chaining rather than one large prompt. Each step feeds the next: identify conditions, apply the Caluza Triangle, check regulatory requirements, spot evidence gaps. That sequence follows the order an experienced attorney works through a case in. We kept tuning token usage to hold costs down as volume grew.

Phase 4: Attorney-facing tools

We built the interface where legal teams review AI outputs, trigger letter generation, and manage cases. Color-coded evidence tables show the status of each condition at a glance. Prompt templates are editable in the app, so non-technical users can change how CARL analyzes a case without waiting on a developer.

Phase 5: Quality assurance

We validated every AI output against actual case outcomes. We also consolidated letter types, improved processing performance, and built QA workflows that check outputs against legal and evidentiary standards before they reach an attorney's desk.