Portfolio case study · B2B SaaS · RAG · WhatsApp
A role-based construction management platform combining live project data, real-time dashboards, approval automation, and a WhatsApp RAG assistant for field teams.
The operating problem
Project status, field questions, documents, and approvals were moving through separate channels. The operating problem was not simply answering questions: teams needed current project context in the channel they already used, while approvals and records remained controlled by the core platform.
The delivered system
The delivered product combined a role-based operations platform, PostgreSQL project data, real-time dashboards, approval automation, and a WhatsApp assistant grounded in project material. The AI layer supported field access to context; deterministic application logic continued to own records, permissions, and approval state.
Operating sequence
This sequence describes the delivered product pattern at a functional level. It does not expose confidential client implementation details.
A field user asks a project question in WhatsApp or reviews an issue in the web product.
The system identifies the project context and retrieves relevant approved information.
The assistant returns a grounded response or routes the exception into the operating workflow.
Approvals and status changes remain recorded in the platform for the responsible team to review.
Delivered capabilities
Production controls
Figures are reported outcomes from the original portfolio material and are not presented as independently audited benchmarks.
Current retrieval engineering is moving beyond a single vector lookup toward richer context, mixed retrieval methods, and evidence that survives real user questions. This project applies the same product principle: put grounded access inside the field channel while keeping the system of record authoritative.
Read the primary source Anthropic: Contextual Retrieval