Cognitive / Multi-Agent AI System
Solynth - AI Economic Intelligence Platform
Solo-built economic intelligence platform for theme discovery/monitoring, combining a user-facing RAG chatbot (SSE streaming + session persistence) with an admin automation stack for ingestion, scheduling, and analysis/report workflows.
- Role
- Solo builder · architecture, implementation, and operational tooling
- Period
- 2024 - Present
- Context
- Personal Project
Solo-built economic intelligence platform for theme discovery/monitoring, combining a user-facing RAG chatbot (SSE streaming + session persistence) with an admin automation stack for ingestion, scheduling, and analysis/report workflows.
What made the system difficult.
Coordinating ingestion, analysis, and long-running generation workflows while keeping the system observable and resilient to intermittent provider/API failures.
How boundaries and flows answered it.
Implemented a server-side RAG pipeline (retrieval→context building→multi-provider response) with DB-backed sessions/messages and SSE streaming. Built node-cron schedulers and automation services for economic/news pipelines and theme analysis workflows in the admin stack.
The path of responsibility through the system.
AutoGenerator
Admin Platform - Content Automation
Responsibilities
- — Unified economic scheduler
- — Theme analysis automation
- — Operational pipelines
AInewsWeb
Service Platform - User Intelligence
Responsibilities
- — RAG chatbot core
- — SSE streaming chat
- — Session persistence
Overview
- Solo-built economic intelligence platform combining a user-facing RAG chatbot service with an admin automation stack.
Purpose & User Flow
Purpose
- Help users quickly understand “what themes matter now” and ask evidence-based questions over ingested sources (news/indicators).
User Flow (Chat)
- Ask a question → retrieve relevant context (vector search) → generate an answer → stream status/output via SSE.
- Conversations are stored per session to support continuity and revisiting.
Key Components
RAG Chatbot Service
- Vector retrieval → context assembly → multi-provider response generation, with session/message persistence (Prisma)
- SSE streaming endpoint for token-level output and status events
Ingestion & Scheduling
- node-cron based unified economic scheduler (event-aware triggering + DB cleanup)
Admin Automation
- Theme analysis/stock discovery scheduler integrating multi-source collectors and orchestration