Back to Projects
Completed Project

Enterprise AI Agent Gateway

Production AI orchestration platform for portal, WhatsApp, and WeChat queries, designed around strict tenant isolation and auditable tool execution.

|
Enterprise AI Agent Gateway
RoleSoftware Engineer
Timeline2026
TeamPrivate platform
Tech Stack8 Technologies

Mission Brief

A private enterprise AI platform for freight and logistics operations, serving portal, WhatsApp, and WeChat queries through controlled agent workflows. Every request is tenant-aware and auditable before it reaches a model or tool, with document RAG, typed freight-domain connectors, scheduled SLA automation, and observability built into the platform.

Key Features

Tenant-Safe Agent Workflows

  • Request Guards: Routed every request through Supabase JWT validation, RBAC, and tenant guards before model or tool execution.
  • LangGraph Control: Built workflows with tool-loop caps and human approval gates.
  • Auditability: Added append-only audit logs, LangSmith tracing, OpenTelemetry, and Sentry.

Freight-Domain Tooling

  • Typed Registry: Created a typed tool registry with more than 15 freight-domain connectors.
  • Controlled Execution: Applied schema validation, role scopes, timeouts, and circuit breakers to tool calls.
  • Multi-Channel Operations: Designed the platform for portal, WhatsApp, and WeChat queries.

Grounded Document Intelligence

  • Document RAG: Delivered document retrieval using Supabase Storage, PDF/DOCX parsing, and pgvector.
  • Tenant Boundaries: Applied mandatory tenant filters to retrieval queries.
  • Grounded Answers: Returned grounded citations with document-based responses.

Automation & Operations

  • Scheduled Workflows: Deployed Temporal scheduling and SLA automation on AWS ECS/Fargate.
  • Operational Impact: Reduced manual status inquiries by about 45%.
  • Responsive Reads: Kept p95 read-heavy latency under three seconds.

Engineering Chronicles

Untrusted Model and Tool Requests

SolutionValidated Supabase JWTs, RBAC permissions, and tenant boundaries before model or tool execution.

Complex Freight-Domain Integrations

SolutionCreated a typed registry of more than 15 connectors with schema validation, scoped access, timeouts, and circuit breakers.

Reliable, Grounded Retrieval

SolutionCombined Supabase Storage, document parsing, pgvector, mandatory tenant filters, and citations for auditable RAG responses.

Installation Instructions

Platform Access

Private platform — access is restricted to authorized organization users.

Technical Case Study

Selected architecture details and implementation decisions are available on request.

Technology Focus

Python, FastAPI, LangGraph, Supabase, Next.js, AWS, Temporal, and Redis.

Operational Scope

Multi-tenant freight and logistics operations across portal, WhatsApp, and WeChat.

Project Access

Live DemoSource Code

Technologies

Python
FastAPI
LangGraph
Supabase
Next.js
AWS
Temporal
Redis

Table of Contents

  • • Mission Brief
  • • Key Features
  • • Engineering Chronicles
  • • Installation Instructions
Back to Projects

More Projects

Multi-Tenant Freight Brokerage & TMS Platform
Completed

Multi-Tenant Freight Brokerage & TMS Platform

Java • Logistics Technology

Shippure
Completed

Shippure

Java 21 • Logistics Technology

FTE Operations Platform
Completed

FTE Operations Platform

Python • Logistics Technology