Accreva Business Ledger
Developer intern work across the Accreva product cycle: frontend app development, integration support, FastAPI domain API, MCP tool layer, and ledger-safe agent workflows.
Backend core
Phase B done
Local architecture docs
AR/AP
Phase C active
Local project plan
Layer
Web
React UI
Layer
API
FastAPI rules
Layer
Postgres
System of record
Layer
Agent
Planner / executor
Layer
MCP
Tool policy
Layer
Audit log
Mutation trace
Problem
A ledger product cannot let model output write business records directly. The app needs reliable product work and deterministic rules before AI enters the loop.
- Double-entry operations need balance checks.
- Posted records need immutability.
- Every tenant-scoped mutation needs an audit trail.
System
The architecture splits UI, API, MCP, agent, and database responsibilities. My intern work supports the product cycle, currently with a frontend-heavy focus.
- React web app calls API and agent services only.
- FastAPI owns business rules and database access.
- MCP tools expose safe ledger operations to agents.
Shipped proof
Accreva is active product work, not my business. The proof is contribution across the repo and a ledger architecture that keeps AI away from raw database writes.
- Developer intern supporting the lead developer across the development cycle.
- Partners, payment terms, invoices, and payments in Phase C.
- Reports hardening and final UX polish remain.
Lesson
Agent systems get safer when AI decides the route and deterministic code owns the mutation.
- Keep private code private.
- Expose public architecture and proof.
- Use confirmation gates for write actions.
Evidence links
Repo stays private until approved.