Cognalytix is the core backend services and AI orchestration engine that turns personal journal entries into structured self-discovery cards. It processes raw journal text, detects emotional patterns across topic families using a local classification sidecar, semantically deduplicates user vocabulary via PostgreSQL pgvector, and narrates growth trends using localized LLMs.
- Dual-AI Architecture: Leverages local ONNX models for sub-second paragraph classification and Ollama chat models for high-level insight synthesis and narration.
-
Semantic Vocabulary Matching: Utilizes
nomic-embed-textand PostgreSQLpgvectorto identify and reuse semantically similar user labels ($\ge 0.75$ cosine similarity), preventing tag duplication. - Explainable Trajectories: Combines SQL-first statistical aggregation with LLM natural language generation to explain emotional shifts.
- Tabular Data Export: Offers a flat, paginated JSON export of journal sections for analytics tools (e.g., Power BI).
The service mesh operates as a containerized stack that connects to a native host-level instance of Ollama to leverage GPU hardware acceleration.
graph TD
Client[React Frontend / Nginx :5173] -->|API Proxy /api| Backend[Spring Boot Backend :8000]
Backend -->|Data & pgvector Embeddings| DB[(PostgreSQL :5332 + pgvector)]
Backend -->|Async Paragraph Classification| Sidecar[Python ONNX Sidecar :8001]
Backend -->|Local LLM & Embeddings| Ollama[Host Ollama :11434]
subgraph Python ONNX Sidecar
Sidecar -->|Roberta| GoEmotions[GoEmotions Classifier]
Sidecar -->|MiniLM| Keyphrase[Keyphrase Extractor]
Sidecar -->|VADER| Polarity[Polarity Intensity]
end
subgraph Host Ollama
Ollama -->|qwen3.5:4b| ChatModel[Summary & Narration]
Ollama -->|nomic-embed-text| EmbedModel[Semantic Label Vectorizer]
end
Explore the following guides for detailed implementation and setup:
- 🚀 Getting Started Guide: Installation steps, host Ollama prep, local development, and port mappings.
- ⚙️ System Architecture: The Dual-AI engine division of labor, label matching, and mirror narration.
- 🔌 API Reference: Complete REST endpoint documentation, request/response bodies, and authentication controls.
- 🗄️ Database Schema: Schema entity details, pgvector index definitions, JSONB shapes, and migrations.
- 🔧 Troubleshooting Guide: Compile fixes, network connection resolutions, and cold-start warmup settings.
To run the backend alongside its database and ONNX sidecar:
- Pull required models on host Ollama:
ollama pull qwen3.5:4b ollama pull nomic-embed-text ollama pull qwen3.5:0.8b
- Start backend and infrastructure:
Ensure you run the Docker compose from the repository root (parent folder):
docker compose up -d postgres sidecar backend
- Verify backend health:
curl -sf http://localhost:8000/actuator/health
Private repository. All rights reserved.