Broker Relationship Management Platform

Confidentiality: Delivered under a consulting engagement. The client's identity and all branded/internal identifiers have been withheld; only my own work, the general architecture, and the technologies used are described.

Overview#

A platform for managing contact information and broker relationships for a company in the agricultural marketing sector. It's a monorepo with a React + TypeScript frontend, an Express + TypeScript API on PostgreSQL, and a full Azure infrastructure footprint defined in Terraform — notable for being delivered under a strict, non-negotiable test-driven development process. A significant part of the product is AI-driven: it transcribes and analyzes phone calls and online meetings, prepares pre-call briefings for brokers, and surfaces trends and alerts across client and market data.

The Problem#

The client needed a secure, maintainable system to manage contacts and the broker relationships connected to them, with enterprise-grade authentication and a deployment pipeline that could safely promote changes across environments.

My Role#

Senior Software Engineer, Tech Lead, and Delivery Lead on the engagement.

Architecture & Approach#

• Frontend. React + TypeScript (Webpack, no framework), React Router, styled-components, and MSAL for Azure AD B2C authentication, with contact directory, detail, and create flows behind protected routes.
• API. Express + TypeScript on PostgreSQL via the Drizzle ORM, with RS256 JWT validation against the B2C JWKS endpoint, pagination/search/filter on list endpoints, and an environment-aware email subsystem (cloud email service in production, file backend locally).
• Infrastructure. A 12-module Terraform setup on Azure — virtual network with public/private/bastion subnets, Container Apps with managed identity, PostgreSQL Flexible Server, Key Vault, Front Door Premium with WAF, plus monitoring, alerting, and policy modules — managed through Terraform Cloud.
• AI features. Two AI-driven capabilities sit alongside the core CRM. The first captures phone calls and online meetings, generates transcripts, analyzes them, and produces pre-call briefings — action items and suggested talking points — ahead of a broker's next conversation. The second runs over broker-client and agricultural commodity-market data (livestock and crops) to detect trends and patterns and raise alerts on the signals that matter.

Technical Highlights#

• Strict TDD as a first-class constraint. Every feature and fix followed an explicit red → green → refactor loop, one test at a time, with production code written only to satisfy a failing test. This was the single most important rule of the codebase.
• Enterprise identity. Azure AD B2C integration spanning the SPA (MSAL) and the API (JWKS-based JWT validation), including custom-branded sign-in and password-reset pages.
• Hardened, automated delivery. An eight-job CI/CD pipeline using Azure OIDC federation (no stored credentials) to build, test, and deploy across QA and staging, with Front Door routing, WAF rate-limiting and OWASP rules, and managed-identity access to the registry and Key Vault.
• AI-assisted broker workflows. Speech-to-text transcription of calls and meetings feeds an analysis-and-summarization layer that turns each conversation into structured pre-call briefings, while a separate analytics layer mines client and market data for trends, patterns, and alerts — all delivered under the same strict TDD discipline as the rest of the system.

Skills Demonstrated#

AI/LLM integration for transcription, analysis, briefing generation, and trend detection; Azure cloud architecture, disciplined test-driven development, enterprise identity and access management, infrastructure-as-code, and secure full-stack delivery — a useful counterpart to my AWS-based work, since this engagement was Azure end to end.