RICCARDO GERMANÀAI ENGINEERING / ENTERPRISE ARCHITECTURE

Developer by instinct. Architect by perspective.

From ideasto systemsthat act.

I’m Riccardo Germanà. I write code, design architectures and lead teams. My terrain is where AI, voice and enterprise systems meet.

Explore my workLet’s build something
SYSTEM STUDY / 01RG — R&D
Generative three-dimensional sculpture of a connected system
One goal. Multiple agents. One conductor.Reason → orchestrate → act
Interactive study · not live telemetry

A manager’s perspective. A builder’s mindset.

Meet the person behind the systems

The big picture.
Hands on the code.

I don’t stop at the diagram. I care about what happens when the system meets the real world.

Riccardo Germanà
Riccardo GermanàMessina, IT ↗

Technical Architect and AI Engineer with 9+ years of experience. I combine enterprise architecture, agentic AI and voice technology to create solutions that make a difference. My strength is the 360° vision: from model to deployment, from protocol to pixel.

Years on enterprise delivery
9+
Salesforce Certifications
14
Architect Credentials
7
Hands-on Associate Manager
BIP
01

Architecture

Distributed, multi-tenant, scalable systems. Built to last.

02

Agentic AI

Autonomous agents with tool calling, RAG, MCP and multi-agent orchestration.

03

Voice & Real-time

AI conversations integrated with telephony, contact centers and enterprise workflows.

04

Enterprise Integration

Salesforce, MuleSoft, contact centers. Systems that talk to each other.

A journey, not a straight line.

  1. 2017–2020

    Full Stack Developer

    Custom web projects with JavaScript, React, Angular, Node.js, Java, .NET. Frontend, backend, REST APIs, and DevOps.

  2. 2020–2022

    Salesforce Developer

    Specialized in Apex, LWC, Flows for enterprise clients. First Salesforce certifications.

  3. 2022–2024

    Technical Architect — BIP

    Scalable CRM architectures, MuleSoft integrations, team coordination. Grew to 14 Salesforce certifications.

  4. 2024–Today

    AI Engineer & Architect

    Agentic systems, voice AI, custom MCP servers. I build AI that acts, not just answers.

AI is interesting.
When it does something.

The model is one component. The product is everything you build around it.

From context to action

INPUT → IMPACT
  1. Context01
  2. Reasoning02
  3. Tools & APIs03
  4. Verification04

Reference pattern · human supervision at critical steps

I design end-to-end agentic systems: agents that reason, call tools, orchestrate flows and make autonomous decisions. Specialized in real-time voice AI, MCP servers and agent fleets with quality gates.

01

Agent Engineering

Single and multi-agent systems with tool calling, function calling, persistent memory. Orchestration with team-lead pattern and manual quality gates.

  • Pydantic AI
  • Vercel AI SDK
  • OpenCode
  • Multi-Agent
02

Real-time Voice AI

Complete voice pipelines: STT, VAD, LLM, TTS, turn-taking and SIP telephony. Contact center and enterprise integrations like Salesforce.

  • ElevenLabs
  • LiveKit
  • Deepgram
  • WebRTC
03

MCP — Model Context Protocol

I build custom MCP servers to expose tools, resources and knowledge bases to LLMs. MCP clients integrated into production agents.

  • MCP SDK
  • Tool Servers
  • Resource Servers
04

RAG & Knowledge

RAG systems with hybrid search (FTS + semantic), pgvector, Qdrant, local embeddings, version-aware retrieval. Knowledge bases that scale.

  • pgvector
  • Qdrant
  • Embeddings
  • Hybrid Search
05

LLM Orchestration

Multi-provider routing (Claude, GPT, Gemini, Grok), structured prompt engineering, automated evals, A/B testing on prompts and models.

  • Claude
  • OpenAI
  • Gemini
  • Routing
06

Production AI Ops

Observability (OpenTelemetry, Sentry), hypothesis-driven test loops, guardrails, cost management, multi-tenant enterprise deploys with full audit.

  • OTel
  • Evals
  • Guardrails
  • Multi-tenant

Fewer promises.
More systems.

Six projects. Concrete decisions, real constraints and code that goes beyond the demo.

Anonymized case studies · NDA-safe

Automotive · national leaderCASE / 01

Enterprise voice agent

70+ locations, voice customer service

System built

A voice AI platform with multi-site SIP routing, knowledge retrieval, guardrails and Salesforce handoff.

My role
Technical Architect & Delivery Lead
The outcome
24/7 service, human handoff preserved, across 70+ locations with centralized cloud orchestration.
  • ElevenLabs
  • Salesforce
  • Qdrant
  • AWS
  • SIP
Inside the project

The problem

Automate voice customer service at national scale without losing quality, keeping human handoff on complex cases and integration with the existing CRM.

Architecture schematic

  1. ChannelsSIP telephony · multi-site routing
  2. Voice platformElevenLabs Conv AI · ASR/TTS
  3. OrchestrationLLM · RAG · guardrails · turn-taking
  4. Knowledge baseQdrant · semantic retrieval
  5. IntegrationsSalesforce Service Cloud · human handoff

The decisions

Direct SIP routing over PSTN; RAG on a knowledge base over fine-tuning; guardrails and human handoff non-negotiable on sensitive cases.

Mobility · national leaderCASE / 02

API-led integration suite

10 service areas, 13 systems

System built

A three-layer MuleSoft API-led suite connecting channels, processes and core systems without propagating legacy coupling.

My role
Integration Architect
The outcome
Suite in production across 10 service areas, legacy systems decoupled, reusable API contracts.
  • MuleSoft
  • CloudHub
  • DataWeave
  • REST
  • RAML
Inside the project

The problem

Integrate loyalty, ERP, ticketing, billing and 10+ legacy systems into a coherent, maintainable architecture, in production.

Architecture schematic

  1. Experience APIWeb · mobile · partner
  2. Process APILoyalty · pricing · orders
  3. System APISalesforce · ERP · legacy
  4. RuntimeMuleSoft CloudHub · DataWeave

The decisions

API-led layering to decouple legacy; RAML contracts as source of truth; CDC for near-realtime sync across domains.

Energy · regulated tariffsCASE / 03

AI automation with quality gates

Document BPO automation

System built

A hybrid document pipeline: deterministic parser, LLM fallback, human validation and versioned output.

My role
AI Engineer & Architect
The outcome
Automation with explicit quality gates, full traceability, aligned with compliance requirements.
  • Python
  • FastAPI
  • Next.js
  • LLM
  • HITL
Inside the project

The problem

Automate parsing and normalization of regulated documents (ARERA/Terna) with high accuracy and full compliance.

Architecture schematic

  1. IngestionRegulated docs · multi-format
  2. ParserContent-driven · deterministic
  3. LLM fallbackUnstructured cases
  4. HITL validationHuman-in-the-loop mandatory
  5. OutputStructured · versioned · auditable

The decisions

Deterministic parser first, LLM only as fallback; mandatory human validation; 55 regression tests as guardrail.

R&D · internal toolingCASE / 04

Agent framework & MCP server

3 MCP servers, 44+ tools

System built

A multi-agent framework and three MCP servers exposing 44+ tools, knowledge bases and enterprise APIs.

My role
AI Engineer · Builder
The outcome
3 MCP servers in production, automation of the enterprise integration lifecycle, automated platform assessments.
  • MCP
  • TypeScript
  • RAG
  • Multi-Agent
Inside the project

The problem

Boost team efficiency with agents that automate analysis, documentation and deployment of enterprise platforms (Salesforce, MuleSoft).

Architecture schematic

  1. MCP serverElevenLabs API · MuleSoft RAG · Knowledge base
  2. Agent frameworkMulti-agent · team-lead pattern
  3. ToolingCLI · IDE integration
  4. Quality gateEvals · manual review

The decisions

Local version-aware RAG (220K+ chunks); entire APIs exposed as LLM tools; manual quality gates before merge.

Side project · personalCASE / 05

CoBrain

Local-first meeting copilot

System built

A local-first copilot for meeting capture, transcription, diarization and search over a personal vault.

My role
Solo developer
The outcome
Transcripts, diarization and semantic search over your own meetings, with data never leaving the device.
  • Python
  • Swift
  • Whisper
  • RAG
  • Local-first
Inside the project

The problem

A copilot for Teams meetings that transcribes, diarizes and indexes into a local knowledge base, without ever sending data to third-party servers.

Architecture schematic

  1. CaptureTeams · ASR streaming
  2. ASR + diarizationWhisper · sherpa-onnx
  3. Knowledge baseLocal RAG · personal vault
  4. ClientSwift · local-first

The decisions

Fully local by principle (privacy-first); sherpa-onnx for on-device diarization; RAG on a personal vault instead of cloud LLMs.

Side project · open sourceCASE / 06

SF Analyzer

Agentic AI for Salesforce orgs

System built

A hierarchical agent fleet that inspects Salesforce orgs and produces documentation behind explicit quality gates.

My role
Creator · Solo developer
The outcome
v1 in production: automated analysis of complex orgs, documentation generated and validated by leader agents.
  • TypeScript
  • React
  • Vercel AI SDK
  • Multi-Agent
  • Salesforce
Inside the project

The problem

Automate the analysis, documentation and assessment of complex Salesforce orgs through a fleet of specialized AI agents.

Architecture schematic

  1. Agent fleet9 documentor · 3 leader
  2. OrchestrationTeam-lead pattern · quality gate
  3. BackendNode.js · TypeScript · PostgreSQL
  4. ClientReact · Vercel AI SDK

The decisions

Team-lead pattern with domain-specialized agents; manual quality gates before merge; v2 pivoting local-first.

The problem first.
The stack second.

Polyglot full-stack: AI, web, mobile, enterprise. I pick the right tool for the problem, not the problem for the tool.

01

AI & LLM

  • Claude
  • OpenAI
  • Gemini
  • Grok
  • Pydantic AI
  • Vercel AI SDK
  • MCP
  • LangChain
02

Voice AI

  • ElevenLabs
  • LiveKit
  • Deepgram
  • Whisper
  • Silero VAD
  • WebRTC
  • SIP
  • Twilio
03

Frontend

  • React
  • Next.js
  • Astro
  • SolidJS
  • TypeScript
  • Tailwind
  • Motion
  • shadcn/ui
04

Backend

  • Node.js
  • Fastify
  • FastAPI
  • NestJS
  • Python
  • Rust
  • Java Spring
  • GraphQL
05

Salesforce

  • Apex
  • LWC
  • Flows
  • Sales Cloud
  • Service Cloud
  • Marketing Cloud
  • Data Cloud
  • Agentforce
06

Integration

  • MuleSoft
  • Anypoint
  • DataWeave
  • RAML
  • REST API
  • SOAP
  • Event-Driven
  • Webhooks
07

Database

  • PostgreSQL
  • pgvector
  • Qdrant
  • MongoDB
  • Redis
  • SQLite
  • Prisma
  • SQLAlchemy
08

DevOps & Cloud

  • Docker
  • Kubernetes
  • AWS
  • Azure
  • CI/CD
  • Terraform
  • OpenTelemetry
  • Linux

Experiment, yes.
Improvise, no.

Speed matters. Knowing what you’re shipping matters more.

  1. 01

    Discovery

    I understand the real problem, not just requirements. Stress-test assumptions, map constraints, define measurable success.

  2. 02

    Architecture

    I design the solution: pattern selection, data model, API contracts, AI strategy. Decisions documented as ADRs.

  3. 03

    Slicing

    Decompose into independent vertical slices (tracer bullets). Each slice delivers end-to-end value and ships on its own.

  4. 04

    Build

    TDD where it makes sense, AI-augmented always. Type-safe end-to-end, observability by default, hypothesis-driven testing.

  5. 05

    Verify

    Automated evals for AI, E2E tests for the rest. Quality gates before every merge. No claims without fresh evidence.

  6. 06

    Ship & Learn

    Feature-flagged deploys, real-time monitoring, structured post-mortems. Every release improves the system and the process.

The best ideas
start with a conversation.

Got an AI idea, a system to architect, an agent to build? Let's talk.

Prefer a direct conversation?

+39 351 830 4323Messina, Italy

Messages are sent through EmailJS. Please don’t include confidential data.