Kofi E. Ampaabeng

Selected Work

20 systems I have architected, built and run: enterprise platforms, consumer apps, and the tools and infrastructure behind them. 20+ years across regulated banking, data platforms and applied ML.

20
systems
6
enterprise
7
ML · LLM
20+
years

09 · AI platform

CweCi gateway

Cape Wesley's self-hosted AI assistant for engineering teams.

A self-hosted, OpenAI-compatible gateway that routes a team's AI through one endpoint: local models first, cloud fallback, and a guardrail pipeline on every request.

  • Tool

Surfaces

  • API
  • Web chat
  • Self-hosted
  1. 01Local first — Ollama by default, with fallback to Claude and GPT models; a new model is configuration, not code.
  2. 02Guardrail pipeline — PII redaction (Presidio), then injection defence, then engineering policy, then audit, in that order.
  3. 03An assistant on top — CweCi, the company website's chat assistant, runs through it with a rate limit and a scoped prompt.
  4. 04Observable — Prometheus and Grafana dashboards for latency, spend and errors.
  5. 05Observability and analytics — Prometheus and Grafana dashboards for model throughput, latency, cost and usage.
  • Local models first
  • Cloud fallback
  • Guardrails in order
SurfacesAPIWeb chatSelf-hosted
CLIENTSEDGEGATEWAYMODELSRECORDfirstfallbackaudit · costmetricsWebsite chatEditors · appsStaff chatOpen WebUIAccess + API keysLiteLLM routerGuardrailsPII · injection · policy · auditPresidioOllama (local)Claude · GPTPostgreSQLkeys · audit · costPrometheus · Grafana
Architecture, drawn from the codeServiceStoreModel usedModel trained hereFeedback
  • LiteLLM
  • Ollama
  • Presidio
  • PostgreSQL · Redis
  • Open WebUI
  • Prometheus · Grafana