# Amalfi AI

> Amalfi AI is a premier consultancy that operationalizes AI governance, AI cost and AI performance — AI GovOps, AI FinOps and AI Performance Engineering — by embedding controls, metering and performance engineering directly into the enterprise AI infrastructure clients already run. Governed, cost-controlled and performant AI — in production, not on paper.

- Website: https://amalfi.ai/
- Founder: John Joseph D'Esposito — https://john.desposito.us/
- Book a 30-minute AI operations review: https://calendly.com/johndesp-amalfi
- Contact: johndesp@amalfi.ai
- Trusted by: Goldman Sachs, United Airlines, Nomi Health, DXC Technology (Fortune 500 and enterprise clients)

## Core competencies

- [AI GovOps](https://amalfi.ai/#govops): Governance as a continuous flow, not a static checklist. Real-time risk classification, automated guardrails and immutable audit evidence generated at the point of every AI interaction. Controls live inside the API gateway, the data pipeline and the deployment workflow. Mapped to ISO 42001, NIST AI RMF, the EU AI Act, HIPAA and SOC 2 through an 88-control AI Governance Hardening Framework across five pillars: Governance & Accountability, Risk Management, Data Integrity & Privacy, Model Lifecycle Security, and Compliance & Audit. Prompt risk classification is delivered by PromptClassify (https://promptclassify.ai), Amalfi AI's classification engine — a 0–100 risk score and regulatory flags for every prompt, running inside the client's own perimeter.
- [AI FinOps](https://amalfi.ai/#finops): Knowing — and controlling — what AI actually costs. Every AI call — LLM, MCP, agent-to-agent — is metered, attributed to an owner (agent, team, model, provider) and reconciled in a ledger finance can match against the invoice. Visibility comes first, measured from live traffic with no application changes — including coding agents such as Claude Code. Then costs come down in-line, on the fly: waste removed, per-team and per-session budgets and hard caps enforced in the request path.
- [AI Performance Engineering](https://amalfi.ai/#performance): Making AI fast and reliable in production. Model-agnostic: routing across frontier, open-weight and local models (for example Gemma running locally). Time to first token, agent turn time and tail latency engineered and held to SLOs. Complexity-aware routing, semantic and prefix caching, capacity balancing, and provider failover keep agents running through outages, rate limits and quota exhaustion.

## Where we build

- AI gateway engineering: Kong AI Gateway, Kong Konnect and the Nginx / OpenResty layer beneath them as the enforcement point — risk-aware LLM routing, prompt classification, spend metering and performance policies as custom Lua plugins; Nginx/OpenResty/Lua tuned as part of AI performance engineering; OWASP LLM Top 10.
- Lakehouse & MLOps: Databricks medallion architecture, MLflow and Unity Catalog as the system of record for audit evidence, the AI cost ledger and performance telemetry.
- Observability: Dynatrace and OpenTelemetry instrumentation — governance SLAs, spend per call and latency per model on one pane, with custom DQL for drift, error-rate and cost anomalies.

## How we engage

Amalfi AI works with a small number of enterprises as a near-partner on their AI operations.

1. Measure: a short baseline from live AI traffic — cost, performance and governance exposure — with no application changes.
2. Engineer: controls, metering and performance policies embedded in the gateway, data pipeline and deployment workflow, built as code and tested like software.
3. Operate: SLOs, budgets and audit evidence reviewed continuously as the AI estate grows.

## Optional

- [PromptClassify](https://promptclassify.ai): Amalfi AI's prompt risk classification engine.