Role: Full Stack Engineer (Solo Founder)
Status: In Development (Private Commercial Repo)
A monetized B2B SaaS platform that uses Retrieval-Augmented Generation (RAG) to provide evidence-based aromatherapy protocols and information about essential oils. The system is designed to handle complex context injection while managing user quotas and subscriptions.
Real-time process for generating context-aware responses during user interaction:
- Semantic Retrieval: User queries are vectorized on-demand (float32) to perform a cosine similarity search against the pgvector database, identifying the most relevant protocols and information.
- Context Injection: The retrieved data nodes and active conversation history are injected into the LLM system prompt as the primary knowledge source.
- Bounded Generation: The model is constrained to generate responses based only on the provided context. This eliminates hallucinations by preventing the LLM from relying on its general training data for technical product information.
- Go & HTMX: Utilized Go for high-concurrency backend logic and HTMX for Server-Side Rendering (SSR), reducing client-side bundle size and complexity.
- Redis Strategy: Implemented Redis not just for sessions, but for real-time usage tracking. This enforces subscription limits (e.g., 100 queries/month) without hitting the primary database for every check.
- Self-hosted on a VPS using Dokploy and Traefik for reverse proxy management and SSL termination.
