Technical Capabilities
Systems Built and Running
Everything listed here is something I designed, built, and currently operate. Nothing aspirational. These systems run daily across a self-managed infrastructure fleet.
Production Systems
Multi-Model LLM Gateway
Multi-model routing gateway with workload-semantic routes. Each route is named by what it does, not which model backs it — so swapping the underlying model is a config change, not a code change. Cross-provider fallback chains span Anthropic, Google, DeepSeek, and Mercury. Deployed on a Hostinger VPS, bound to a Tailscale mesh interface.
Semantic Memory System
Long-term memory store holding 1,800+ memories in PostgreSQL with pgvector (768-dimensional embeddings via nomic-embed-text through Ollama). Self-hosted on a Synology NAS with a Supabase warm standby for failover. Exposed as an MCP server for AI coding agents and chat interfaces. Supports keyword search, semantic search, topic mapping, and memory lifecycle management.
Nightly Content-Intelligence Pipeline
Automated content capture, synthesis, and digest delivery. Seven scheduled nightly jobs handle ingestion from multiple platforms, content summarization, connection discovery between stored memories, and morning briefing generation. Built on Docker with Ollama for local inference.
Self-Hosted Infrastructure Fleet
Multi-node fleet running 24/7. Dual RTX 5060 Ti workstation for GPU inference, two Apple M4 Mac Minis for application services, two Synology NAS units for storage and database hosting, and a Hostinger VPS for the gateway. Connected via Tailscale mesh VPN, with Docker containers across nodes behind an nginx reverse proxy.
Practice-Management Platform (ConsultRussRE)
Modular practice-management system for real estate operations. Phase 1 (relationship-maintenance module) built with full test coverage — 15 passing tests; additional modules designed. Deployment pending.
Open-Source Reference Implementations
Public, generic reference implementations distilled from systems I operate privately, each shipping its own eval suite:
memstrata (semantic-memory MCP server),
tiergate (privacy-tier LLM routing — data classification, not cost, picks the model), and
paperjudge (document-compliance pipeline). More at
github.com/ConsultRuss.
Technical Stack
- LLM orchestration — LiteLLM, Ollama, multi-provider API integration
- MCP (Model Context Protocol) server development
- PostgreSQL 16 with pgvector for vector similarity search
- Supabase (BaaS, Edge Functions, warm standby)
- Docker containerization across Linux and Apple Silicon
- Tailscale mesh VPN for multi-node connectivity
- nginx reverse proxy configuration
- Synology NAS administration
- Python, Node.js, SQL
- Claude skill and prompt engineering