Exponent — AI/DevOps Automation Platform

Founded Exponential and built Exponent, an AI agent orchestration platform for infrastructure and development automation — built from scratch and running in production on a Hetzner VPS (systemd).

Highlights

• Agents: LangGraph agent with persistent state — externalised checkpoints in Postgres, vector memory in Qdrant — plus a SmolAgent fallback with local inference via llama.cpp behind an OpenAI-compatible endpoint.
• Architecture: Docker microservices — FastAPI (Python) API, React frontend, PostgreSQL, Redis and n8n workflow automation; MCP integration for container management and infrastructure operations; Terraform-managed infrastructure, health probes and graceful shutdown (SIGTERM draining).
• Full-stack product: React dashboard, FastAPI backend (~200 endpoints), Flutter iOS/Android app, and an embeddable chatbot widget with session persistence, rate limiting and domain allowlisting.
• RAG, memory & retrieval: WebRAG hybrid search over an internal knowledge base plus Tavily web search, per-user conversation memory, and prompt/response caching — ~90% LLM cost reduction on repeat workloads.
• Extensibility & governance: plugin marketplace for extensibility; audit trail aligned with ISO/IEC 42001.
• AI-assisted delivery & testing: day-to-day pair-programming with Claude Code, Cursor and Codex; unit suites with coverage thresholds, integration tests against live services and TDD on the orchestration core, all enforced through CI before deploy.
• Infra: Kubernetes with custom CRDs for agent pods, Helm, Kustomize and Terraform across Docker / AWS / GCP / Azure, with Ansible for post-deploy config.

Stack: Python, FastAPI, LangGraph, LangChain, SmolAgents, llama.cpp, Ollama, Qdrant, Tavily, React, Flutter, PostgreSQL, Redis, n8n, Docker, Kubernetes, Helm, Terraform, GitHub Actions.