Connect IQ.
A centralized networking platform powered by AI to facilitate intelligent professional connections and encrypted communication.
The Challenge
The core mandate was to build a professional networking platform that transcended conventional keyword-based matching by leveraging advanced natural language understanding. The system needed to process unstructured professional data — resumes, project portfolios, endorsement narratives — and distill it into semantically meaningful vectors for high-precision matching. A critical constraint was end-to-end encryption: all communication channels had to be secured such that even at the infrastructure level, message content remained inaccessible. The AI inference pipeline posed a significant engineering challenge — OpenAI API calls needed to return sub-second responses while handling hundreds of concurrent user sessions, requiring careful orchestration between the Node.js API gateway and Python-based ML microservices. Additionally, the platform's recommendation engine had to balance relevance with diversity, preventing filter bubbles while surfacing genuinely valuable professional connections across disparate industries.
The Solution
We engineered a multi-tiered architecture with Next.js handling the server-rendered frontend and API routes, a Node.js backend managing session orchestration and real-time messaging via WebSockets, and a containerized Python microservice dedicated to AI workloads. Professional profiles were processed through OpenAI's embedding models to generate high-dimensional semantic vectors, stored in an optimized vector database with HNSW indexing for nearest-neighbor searches completing in under 50ms. The encryption layer implemented the Signal Protocol for end-to-end encrypted messaging, with key exchange handled through a zero-knowledge architecture. An async task queue (backed by Redis) decoupled AI inference from the request-response cycle, allowing the Node.js layer to serve users non-blockingly while Python workers processed heavy NLP operations. The recommendation engine employed a hybrid collaborative-content filtering model with diversity injection, delivering match relevance scores exceeding 87% in user satisfaction surveys.