Professional work · AI systems · 2025 — 2026
500 Acres AI Support System
A production support platform that retrieves answers from approved company knowledge and gives staff control over configuration, review, and publishing.
- Role
- Design + engineering
- Timeline
- 2025 — 2026
- Stack
- Next.js · TypeScript · Postgres + pgvector · OpenAI APIs · Jest · Vercel
- Metrics
- 2 Next.js apps · 41 test files · Multi-instance control plane
Overview
Scout grew from a set of GPT-based helpers into a multi-instance runtime and control plane for chat, knowledge retrieval, configuration, and operational review.
Policies and operating answers lived across disconnected sources. Staff needed reliable retrieval and a secure way to update knowledge, inspect failures, and change assistant behavior without code deployments.
What I built
Built the Next.js chat runtime and a separate authenticated dashboard that proxies signed requests to control APIs.
Implemented Postgres-backed configuration, pgvector knowledge retrieval, document ingestion, and review and publishing workflows.
Added health checks, integration diagnostics, production preflight scripts, Jest coverage, and documented migration and rollback procedures.
Key choices
Separated the public runtime from the authenticated control plane so operational access and customer-facing traffic have distinct boundaries.
Stored per-instance behavior in Postgres while keeping infrastructure secrets in deployment configuration.
Added payload limits, rate limits, SSRF protections, retention controls, and encrypted storage for sensitive connector values.
Outcomes
Provides one reusable platform for multiple company instances, each with isolated configuration, knowledge, roles, and integrations.
Lets authorized staff test connections, review unresolved questions, manage knowledge, and adjust runtime behavior from the dashboard.
Supports production readiness checks through linting, builds, automated tests, health endpoints, and deployment smoke scripts.