Work
PhantomPulse
A white-label, multi-tenant marketing CRM with AI agents, RAG, and an AI content studio. Built at Valence Tech.
Info
- Client
- Valence Tech · NDA client
- Role
- AI service and UI redesign
- Timeline
- Jul 2026 – present
- Year
- 2026

Overview
A white-label marketing and CRM platform for agencies. An agency resells the product under its own brand, and each of its clients gets an isolated sub-account with CRM, a unified inbox, campaigns, workflow automation, social scheduling, reviews, and AI tools.
My role
At Valence Tech, I built the Python AI service alone and did the product UI redesign alone. I also built backend modules in the .NET modular monolith and the React screens that use them.
Architecture
The core is an ASP.NET Core modular monolith with eight modules on PostgreSQL, Hangfire for background jobs, and SignalR on a Redis backplane for real-time events.
The AI service is a separate FastAPI app. One model router sits over OpenAI, Gemini, and Anthropic, and model pricing is data, so a price change needs no deploy. Each tenant has its own RAG knowledge base on pgvector.
Solution
AI inbox agents make one structured LLM call that returns the reply, the CRM actions, a handoff decision, and a confidence score. A deterministic guardrail hands the conversation to a human when the agent should not answer.
Every AI call goes through a credit wallet: the service reserves credits, runs the call, then settles the real cost with the billing module.
Challenges
Tenant isolation. Data is split by agency and sub-account in both services, the AI service runs on a database role that can only read and write rows, and a dedicated test project checks that no tenant can see another tenant's data.
Stack
- ASP.NET Core
- .NET 10
- FastAPI
- LangGraph
- pgvector
- PostgreSQL
- Hangfire
- SignalR
- React
- AWS
Next project
ValenceTech
A full redesign and rebuild of the site for a studio that ships web, mobile, and AI products.
View case study