Industry
Professional Services
Case Study
Client Operations Workspace
Client context gets lost across inboxes, folders, and personal task lists. I am building one operations platform where deterministic workflows keep work moving and AI helps with extraction and drafting without taking final decisions away from people.
Project Overview
Professional services teams handle parallel conversations, documents, deadlines, approvals, and follow-ups. When each part lives in a different tool, staff repeatedly reconstruct context and important handoffs become fragile.
I am building Client360 as a private operations and CRM platform for this work. It is still in development, so the case study describes implemented foundations rather than presenting the system as a finished production product.
An in-development operations platform that unifies client records, documents, deadlines, durable workflows, and human-reviewed AI assistance.
Services
- UX/UI Design
- Web Development
- Product Strategy
Key Deliverables
- Responsive Design
- Custom Development
- User Experience Improvements
Platform Type
Web Platform
Tech Stack
- TypeScript
- Next.js
- Hono
- tRPC
- PostgreSQL
The Challenge
The system needs to ingest information from email, messaging, and documents, connect it to the correct client, preserve an audit trail, and route the next action to the right person.
AI can reduce repetitive reading and drafting, but deadlines, permissions, workflow transitions, and external communication cannot depend on an unpredictable model response.
Approach
I separated the deterministic business engine from the AI layer. Explicit rules control deadlines, routing, permissions, and state transitions. AI providers can extract context or prepare a draft, but reviewed actions stay behind human approval.
The monorepo separates web, API, database, providers, event storage, projections, mail, observability, and evaluation concerns so integrations can change without rewriting the workflow core.
The Solution
The implemented foundation uses Gmail and Make for intake, a native WhatsApp webhook, Google Drive, and Google Document AI OCR. Durable work is backed by PostgreSQL, Drizzle, pgmq queues, and pg_cron, with row-level security, audit infrastructure, and observable services.
The web application exposes client records, documents, tasks, deadlines, ownership, and reviewed communication in one workspace. OpenAI-compatible providers and a local Ollama proof of concept support assisted work without becoming the source of business truth.
Key Challenges
-
keeping deterministic business rules separate from probabilistic AI output
-
preserving idempotency and auditability across asynchronous document and message intake
-
enforcing role-based access while retaining enough context for useful assistance
-
designing integrations as replaceable providers instead of coupling them to workflow logic
The Outcome
The project now has a credible technical foundation for client operations: typed contracts, durable intake, explicit workflow ownership, auditable state changes, and human-reviewed assistance.
It remains under active development. The current runtime stays focused on document and message workflows; planned voice interaction work and previously evaluated, unused AI UI dependencies are outside the implemented scope.
Context
The central engineering decision was simple: use AI where ambiguity is useful, and deterministic code where the business needs a guarantee.
Related case studies
Other projects with related architecture decisions, platforms, or product challenges.
Greenland Komarov Management
A web-and-mobile operations platform with separate manager and worker experiences, a dedicated API, role-based access, task evidence, and push reminders.
TypeScript · Next.js · Hono · React Native · Expo · PostgreSQL
View case study →ETF Investment Calculator
A multilingual mobile financial-planning toolkit for investment, goal, withdrawal, FIRE, pension, inflation, fee, loan, lump-sum, and real-estate scenarios.
TypeScript · React Native · Expo · Expo Router
View case study →