Data Engineer

Christophe Blefari Portfolio

Co-founder, nao Labs

Blog by Christophe Blefari, a data engineer for 10 years before co-founding nao Labs, known for the "Data News" roundup — weekly from 2022, sent to 7k people at its peak per the site, and sporadic since 2025 once the startup took over.

Data EngineeringData PlatformsNewsletterIndustry Analysis

Location

Paris

Companies

nao Labs

Skills

Data EngineeringData PlatformsNewsletterIndustry Analysis

Notable

  • 10 years in data engineering before co-founding nao Labs (per the About page)
  • Data News newsletter went to 7k people every week, per the site
  • nao, the open-source analytics agent from nao Labs: 1.7k GitHub stars
  • yato, a DuckDB SQL orchestrator: 350 GitHub stars

What makes it work

A breakdown of the choices that make this portfolio stand out.

1

A recurring weekly format — and an honest note when it stalled

Numbered weeks (24.16, 24.20, 24.24 ...) gave the blog a predictable cadence through 2024. The archive also shows what the format costs: 2025 dropped to six posts once nao Labs took over, and the March 2026 "Re." post says so openly instead of letting the archive go quiet.

2

Original commentary alongside curated links

Posts like the Databricks/Snowflake analysis add independent point-of-view on top of the curation, rather than just aggregating headlines — the difference between a newsletter and a genuine blog.

3

Coverage grounded in operational data-team experience

"How to build a data team" and the May 2026 GitHub PR attack writeup (seven malicious PRs that exfiltrated secrets through nao's PR-preview CI, plus which keys were rotated) come from someone running real infrastructure, not observing the space from outside.

What Data Engineers can take from this

Specific, actionable tips to apply to your own portfolio — no generic advice.

  • A recurring, dated format (weekly news, monthly roundup) is easier to sustain long-term than sporadic long-form essays, and it still shows real depth if you add your own take each time.

  • Don't just aggregate links — add a paragraph of independent analysis on why something matters to differentiate curation from commentary.

  • Write from your own operational experience (an incident you handled, a team you built) rather than only commenting on other people's announcements.

Projects

Christophe Blefari's archive is honest about its rhythm: dozens of numbered Data News weeks, then a gap, then a post titled "Re." explaining that a startup took the writing time. The projects below are what went into that gap.

SaaS ProductadvancedOngoing (company product)

nao — open-source analytics agent

TypeScriptPythonCLILLM agentsSQL warehousesdbt

An open-source analytics agent built around "context engineering": `nao init` scaffolds the agent's context as a file system (database schemas, column profiles, docs, example queries, dbt/Looker repos, rules), `nao sync` pulls it from existing sources, `nao test` scores question-to-SQL unit tests, and `nao chat` deploys a chat UI for anyone to query the warehouse.

It treats LLM reliability as a data engineering problem — versioned context, measurable test pass rates — rather than a prompt-writing one.

  • Context lives as plain files (columns.md, profiling.md, RULES.md) that can be versioned and reviewed

  • Built-in test runner reports pass rate, answer rate, latency and token usage

  • Per-PR preview deployments, hardened after the attack writeup

Open SourceintermediateWeekend-sized library

yato — yet another transformation orchestrator

PythonDuckDBSQLdlt

"The smallest orchestrator on Earth" for SQL transformations on DuckDB: point it at a folder of SQL files and it infers the DAG from the queries and runs them in order, each file becoming a table. It works from Python or a CLI (`yato run --db tmp.duckdb sql/`) and is designed to pair with dlt for loading.

Dependency inference from SQL is the core idea behind dbt; rebuilding it this small shows an understanding of the mechanism, not only the tool.

  • No config file — the DAG is guessed from the SQL itself

  • Runs in-memory or against a DuckDB file

ResearchintermediateWeekly, 2022–2025

Data News

NewsletterIndustry analysis

A numbered weekly roundup of the data engineering and AI-tooling ecosystem, mixing curated links with standalone analysis such as "Databricks, Snowflake and the future" (June 2024), which compared the two companies' conferences, revenue and positioning.

Sustaining a weekly, opinionated industry digest for years is visible evidence of reading widely and forming views, which is what senior data roles are paid for.

  • Dated week numbers make the archive easy to browse

  • Standalone analyses sit alongside the link roundups

Ready to build your portfolio?

Follow the Data Engineer roadmap — skills, projects, and timeline to get hired.

Data Engineer Roadmap