Back to Projects
April 2025 - August 2026 • Full Stack

AI Agent Platform (Chat, Integrations & Observability)

Full-stack engineer leading a product pivot from a legacy engagement/incentives platform to a unified AI agent platform, spanning a Next.js frontend, a from-scratch design system, and Go backend services for tool integrations, billing, and LLM usage observability.

B2B SaaS platform that evolved from employee engagement and recognition into a unified AI agent platform for workplace collaboration, tool integrations, and usage analytics

Project Overview

Served as a full-stack engineer at a B2B SaaS company through a major product pivot: from a legacy employee-engagement/incentives platform to a unified AI agent platform for workplace collaboration, tool integrations, and usage analytics. Started as a frontend-focused engineer and expanded into backend ownership across two Go services, eventually solo-designing and deploying a standalone production service.

The Product

The platform centers on an AI agent that helps teams get work done through:

  • Unified Chat: A single chat surface for conversational AI work, with reasoning-step visibility, deep-work/search modes, and message feedback
  • Tool Integrations: OAuth-based connections to third-party systems (CRM, data warehouse, payments) so the agent can retrieve data and take action
  • Canvas: A rich-text collaborative document surface with autosave and resizable layout
  • Performance Reviews: A structured review workflow with progress tracking and self-review flows
  • Billing: Usage-based billing with proactive alerts, automatic overage handling, and admin-managed requests
  • Onboarding: A first-time-user wizard that personalizes agent configuration from user input
  • Observability: Admin-facing dashboards for usage and audit trails of agent activity

Technical Architecture

Frontend (Next.js Monorepo)

  • Framework: Next.js with React and TypeScript
  • Styling: Tailwind CSS, with a from-scratch design system (tokens, Storybook, component library) built and rolled out across the product
  • Data Fetching: React Query; State: Zustand
  • Editor: TipTap-based canvas/rich-text editing with autosave

Backend (Go Services)

  • Primary chat/agent backend: Streaming chat (REST + WebSocket), tool retrieval, billing, and LLM usage observability/audit APIs
  • Legacy backend (in sunset): Maintained its schema layer while business logic migrated to the primary backend
  • Standalone proxy service: Sits in front of an LLM provider’s API, resolving user identity and extracting usage metrics from AI-assistant traffic without altering the underlying requests/responses, deployed on Google Cloud Run

Key Contributions

Product Pivot & Legacy Cleanup

Led frontend delivery as the product moved away from its original employee-engagement model toward the new AI agent platform, culminating in a large-scale cleanup removing a legacy subsystem that no longer fit the new architecture.

Design System

Built a company-wide design system from the ground up - colors, typography, buttons, modals, cards, tabs, forms, badges, toasts, loaders, alerts, and side drawers, documented in Storybook - then migrated existing product surfaces onto it.

Tool Integrations & Billing

Built OAuth-based third-party tool integrations (CRM, data warehouse, payment providers) spanning the frontend and a Go backend, plus a full usage-based billing system: alerts, automatic overage handling, in-app requests, and admin approval workflows.

LLM Observability & Proxy

Built LLM usage observability and audit APIs in Go (per-user/per-network attribution, audit logs with date-range filtering and topic classification), then solo-designed, built, and deployed a production LLM usage-metering proxy service - from initial feasibility spike through cloud deployment.

Onboarding & Customer Observability

Shipped an FTUX onboarding wizard that extracts brand “design DNA” from user input to personalize agent configuration, and a customer-facing usage observability dashboard with audit and topic-trend views - the most recent work delivered before wrapping up the engagement.

Technologies

Next.js React TypeScript Tailwind CSS React Query Zustand Storybook Go PostgreSQL Prisma Redis Anthropic API Composio Stripe Google Cloud Run Docker Terraform WebSocket

Key Highlights

  • Led frontend delivery through a full product pivot from a legacy engagement platform to a unified AI agent chat platform, including a large-scale cleanup removing a legacy subsystem that no longer fit the new architecture
  • Designed and built a company-wide design system from scratch (tokens, Storybook, 15+ core components) and migrated existing product UI onto it
  • Shipped a structured performance-review module end-to-end and an in-app usage-based billing system with proactive alerts and admin approval workflows
  • Built third-party tool integrations (CRM, data warehouse, payments) enabling the AI agent to connect to and act on external systems, spanning frontend and Go backend
  • Extended into backend ownership: built LLM usage observability and audit APIs, and solo-built a production LLM proxy service for usage attribution and cost tracking, deployed on Google Cloud Run
  • Delivered a first-time-user onboarding wizard that auto-generates brand-aware agent configuration from user-supplied inputs