nyami.fr

Ronald Nyami

📍 Switzerland · personal knowledge lab

I document and share responsible GenAI architectures.

10+ years across data, machine learning and architecture. I share practical patterns for GenAI, agents, evaluation, governance and delivery in regulated environments.

AI Architect · GenAI Data Scientist · Banking AI Delivery & Technical Leadership · Medium ↗

10+
years · data, ML & architecture
7
resources · independent or public
1
living agentic-AI demo · this portfolio
Nyami — next-generation AI solutions, quantum-inspired research, end-to-end AI solutions, generative AI and machine learning

New · Interactive Pattern Lab

A RAG demo is 20 lines. A trustworthy AI system is 20 decisions.

An independent, reusable map of the choices that actually matter: data access, retrieval, MCP tools, agent orchestration, evaluations, guardrails and operations.

RAG MCP Agentic AI LLM evaluation LLMOps
Read the technical note↗

Shared knowledge

Public projects and resources

01

in-development

Dataiku Platform Blueprint

A practical reference for environment separation, deployment standards, monitoring and support

→ A reusable way to reason about reliable shared data-science platforms, governance boundaries and operating responsibilities.

Reference architecture diagram — Dataiku Platform Blueprint
architecture — reference
  • dev · test · prod pattern environments
  • by design separation
  • monitoring + runbooks operations
  • Dataiku
  • Kubernetes
  • Python
  • Monitoring

02

in-development

Copilot Studio Blueprint

A secure conversational-app pattern using controlled ingestion, authentication and maintainable workflows

→ A clear framework for reasoning about authentication, content ingestion, governance and maintainability in conversational applications.

Reference architecture diagram — Copilot Studio Blueprint
Request access
  • explicit boundary identity
  • controlled pattern ingestion
  • maintainable workflows
  • Copilot Studio
  • Power Automate
  • Identity controls
  • Controlled ingestion
  • Dataverse

03

paper trading

Controlled Agentic Investment Research

Multi-agent research and paper-trading validation behind deterministic risk, approval and audit controls

→ Split architecture agents / risk engine — hard kill switch · paper validation before real capital.

Reference architecture diagram — Controlled Agentic Investment Research
architecture — reference
  • v1.5.0 shipped
  • 608 tests
  • 95 % coverage
  • Python 3.*
  • LangChain / LangGraph
  • FastAPI
  • Next.js
  • Supabase

04

live

Cost Observability (this site)

How I run agentic AI on a $5/day budget — with a public kill switch

→ AI budget controlled to the cent: per-visitor / per-session / site caps, hard kill switch, public costs on /eval.

Reference architecture diagram — Cost Observability (this site)
architecture — reference
  • $5/day site budget
  • $0.03/day visitor cap
  • ~$0.0003 cost/answer
  • Cloudflare Workers + KV
  • TypeScript
  • SSE streaming

05

delivered

Base44 → Supabase migration

Rebuild of a multi-office EU platform, zero downtime

→ Migration without downtime + removal of the proprietary-vendor lock-in — lower operating cost and full control of the stack regained.

Reference architecture diagram — Base44 → Supabase migration
architecture — reference
  • 113 Base44 refs → 0
  • 8 clean commits
  • 7 tables + RLS
  • Vite + React
  • Supabase (Postgres + RLS)
  • Edge Functions
  • Cloudflare Pages
  • Resend

06

production

Automated video pipeline

100+ shorts produced automatically, 4 platforms

→ 100+ shorts produced with no manual step — automated multi-platform pipeline, cadence held without a human operator.

Reference architecture diagram — Automated video pipeline
Request access
  • 100+ shorts produced
  • ×4 platforms
  • ✓ karaoke captions
  • Python
  • Node.js
  • ffmpeg
  • ElevenLabs
  • Chatterbox

07

published

20 decisions beyond a RAG demo

An interactive pattern map for governed GenAI systems

→ Turns a list of AI tools into an inspectable architecture with explicit controls, evaluation gates and operating boundaries.

Reference architecture diagram — 20 decisions beyond a RAG demo
architecture — reference
  • 20 architecture decisions
  • 5 control planes
  • 0 employer claims
  • RAG + MCP
  • Agentic AI
  • LLM evaluation
  • Kubernetes + LLMOps

Capabilities

Capabilities

Areas explored and technologies used in personal and public projects.

Applied AI Architecture
GenAI application architecture RAG and agentic workflows AI gateway and model abstraction Secure tool and API integration Human review and deterministic controls Evaluation and production readiness
Prototyping & Operability
Python and FastAPI services Docker and Kubernetes Helm and Argo CD CI/CD and GitOps Identity, access and secrets Monitoring and operational practices
Governance & Controls
Use-case risk classification Model approval and lifecycle Access control and auditability Responsible AI control patterns Incident response and rollback Cost and consumption governance
Technical Knowledge Sharing
Architecture facilitation Design reviews Reusable patterns Technical decision records Accessible documentation Knowledge transfer

Technical practice

Python FastAPI LangChain / LangGraph Copilot Studio Power Automate Azure AWS Databricks Dataiku Docker Kubernetes Helm Argo CD PostgreSQL

Architecture and applied research

Power Platform governance Multi-LLM routing MCP Semantic Kernel Prompt Flow LLM evaluation frameworks OpenTelemetry patterns

About

Who I am

Ronald Nyami

Ronald Nyami — AI Architect & GenAI Data Scientist

📍 Switzerland

GenAI architecture notes from ingestion to evaluation — shared as independent educational resources.

« Making complex AI patterns understandable, testable and useful to the community. »

GenAI Data Scientist and AI Architect based in Switzerland, with experience across banking, financial services and technology consulting.

The content covers reference architectures, independent experiments, open-source tools and generic engineering patterns.

The goal is educational: making architecture decisions, limitations, tests and trade-offs useful to the community.

This is a personal, independent site with no commercial or consulting offer. It does not represent any employer or client and publishes no proprietary code, documents, data or architecture. Examples are personal, public or generic. Built with Astryx (Meta, open source).

Outside work: alpine hiking (Switzerland · France · Italy), checkers, football.

Astryx v0.1.7 · Built for people and agents GitHub ↗ Storybook ↗ Sandbox ↗ Docs ↗

Professional profile

AI architecture grounded in a decade of delivery

GenAI Data Scientist and AI Architect based in Switzerland, with experience across private banking, financial services and technology consulting. I translate business needs into secure, scalable AI designs and explain the trade-offs clearly to technical and business stakeholders.

Focus
GenAI · NLP · machine learning · AI architecture
Delivery
End-to-end design · technical leadership · stakeholder alignment
Environment
Banking · regulated settings · confidentiality-aware delivery
Platforms
Dataiku DSS · Kubernetes · MLOps · cloud-native systems

Education & languages

Master’s degrees in Data Science, Big Data & Business Intelligence — École Centrale Paris and Université libre de Bruxelles. English and French: native or bilingual; German: professional working proficiency; Italian: full professional proficiency.

Read the production-grade RAG article on Medium ↗

Values

What I stand for

Evidence over claims

Clear status, traceable decisions and measurable operational results.

Secure by design

Identity, auditability, least privilege and controlled deployment.

Operable AI

Evaluation, telemetry, cost control, incident response and rollback.

Reusable architecture

Patterns and platforms other teams can understand and maintain.

Community

Exchange knowledge

For discussion of a public article or pattern. No professional or commercial service is offered through this site.

Live operations

Agentic Portfolio — Live Operations

Real site telemetry (public endpoints, refreshed 60s). Budgets and kill switch = persistent global data.

GOVERNED

Agentic Systems

EVALUATED

Secure. Scalable. Reliable.

OPERABLE

Measure. Learn. Improve.

TRANSPARENT

Statuses, evidence and roadmaps disclosed.