smiro
for recruiters & hiring managers

I don't apply
for jobs.
I build a system
to find them.

While other candidates refresh job boards and miss windows, my n8n+AI pipeline sees a vacancy 4–8 minutes after it's posted, filters it, drafts the application, and pings me on Telegram.
You're reading this because the system worked.

avg. detection~6 min
platforms3
years building15
based inDijon
01

Full-stack at AI speed

I prototype with Claude Code & Cursor and harden with 7 years in IT. Frontend, backend, infrastructure — idea to production in days, not sprints.

Claude CodeCursorTypeScriptReactNodePython
02

AI agents

Agentic systems that do real work — not chatbots. An executor performs, a reviewer scores, the loop runs until ≥90% quality. RAG, MCP, evals, fallbacks.

ClaudeOpenAIRAGMCPFine-tuning
03

n8n automation

Self-hosted n8n on a VPS I provision, deploy and harden myself. Workflows that run 24/7, integrate 50+ services, and replace 90% of manual ops.

n8nVPSDockerOWASPWebhooks
/01case study · my job alerts

How I get to the inbox first.

A live system I built and run for myself. It's also the proof that what I describe in pillar 02–03 actually works in production. Here's the entire pipeline.

My Job Alerts · workflow

source ai agent filter database output
0trigger
1sources
2scrape
3filter
4store
5act
swipe to explore the pipeline
live · a job moving through
at sources · LinkedIn
~6 min
avg. detection
3×
platforms unified
0
manual scrolling
24/7
running on my VPS
/02open source · github

Dépôts GitHub

The work I keep on top of my GitHub — AI products, full-stack apps and the platform code behind them. Live from @SergeMiro.

01
client site · back-office

ASCoFaçade

Site and back-office of a façade contractor — local SEO up front, four role portals behind the login: office, client, sub-contractor, crew.

Next.js 16React 19TypeScriptTailwind 4daisyUIFramer MotionSupabase AuthPostgreSQLRow-Level SecurityServer ActionsZodreact-hook-formpdf-libjsPDFUppy + tusexifrPannellumRechartsLeafletGoogle MapsGoogle PlacesResendTelegram Bot APIMistralGeminiRAGPWAVercel CronBiome
TypeScript
02
telephony · internal platform

Extranet Fimainfo

Internal telephony platform at Fimainfo — DID numbers and orders, IVR strategies, campaigns, SMS/WhatsApp, BI and an AI assistant. Five services.

Next.js 15.5React 19Tailwind v4shadcn/uiRadix UIZustandZodjose JWTRBACExpress 4Sequelize 6SQL Server ×4FastAPIPythonQdrantDoclingRAG + ACLMistralEvidence.devPinoOpenTelemetrySigNozBrevoMS TeamsWebRTCPlaywrightLadleJestBrunoDockerTraefikGitLab CI
JavaScript
03
Telegram Mini App · loyalty

way2buy

The buyers' club that sells through Telegram, not a website — cashback wallet, Silver/Gold/Platinum tiers, birthday and holiday discounts that fire themselves, a fitting-room cart, a two-channel feed, margin tracking and an admin office with AI reports. Publishing goes app → channel, the webhook brings channel → app. Zero build step, 19 tables, 81 tests.

Node.js 20Express 4SQLitebetter-sqlite319 tables · 15 indexesTelegram Mini App SDKTelegram Bot APIchannel_post webhooklong-poll fallbackvanilla JSzero buildCSS design tokenscashback enginegamificationrules engineidempotent schedulermargin trackingGemini 1.5 FlashAI propose → applynode:test ×81Vercel serverless
TypeScript
04
Automobiles · Mobilités · Conseil

AMC Motors 21

Used-car dealer, front and back — 360° interiors, an AI matchmaker for buyers, and a CRM that turns an enquiry into a signed quote.

Next.jsReact 19TypeScriptTailwindRadix UI ×21TanStack QueryTanStack VirtualSupabasePostgreSQLRow-Level SecurityVercel AI SDKOpenAIVIN decodeBullMQRedisThree.jsReact Three Fiberphoto-sphere-viewerTipTapPDFKitEChartsRechartsnext-intl ×6next-pwa@vercel/ogCheerioPlaywrightVitestBiomeDockerGitHub Actionsn8n
TypeScript
05
AI · compliance · EU back-office

SimplifyEU

AI solutions for EU business — LLM-powered workflows that automate multilingual compliance and back-office paperwork.

Next.js 16React 19TypeScriptTailwind 4Radix UInext-intl ×3StripeTurborepopnpm workspacesFastAPI ×4PythonPydantic v2CrewAI ×5 agentsCeleryRedisAPSchedulerpandaspdfplumberPyPDF2pytesseractPaddleOCRfuzzywuzzyreportlabopenpyxlTwilioslowapiSupabasePostgreSQLDockerTraefikVercelGitHub Actions
Python
06
portfolio · AI avatar

smiro.dev

This site — a static Astro build with no framework on the page, a 3D laptop, and an avatar that answers from one editable Markdown brief.

Astro 6TypeScriptTailwind 4LightningCSSzero hydrationThree.jsWebGLCSS 3DOKLCH themingVercel EdgeSSE streamingOpenRouterGroqGemini 2.5RAG briefDeepgram AuraWeb SpeechGitHub APIhand-rolled i18nMutationObserverSharpn8n
TypeScript
07
AI agents · back-office

Vistalid

Multi-agent back-office for FR call-centre campaigns. Executor works, reviewer grades, loop runs until score ≥ 90%. 30+ hrs/week saved per team. Yes, including URSSAF.

Next.jsTypeScriptClaude APIexecutor + reviewer loopscore gate ≥ 90%PostgreSQLn8nDocker
production · v3.2
08
AI · documents · n8n

Parser AI

Inbox → structured JSON. Executor extracts, reviewer checks, n8n routes. Eats messy emails for breakfast.

Node.jsOpenAI APIexecutor + reviewerper-field rulesgpt-3-encoderfrozen sample setparser_config tablemulti-format intaken8nSQL
production
09
reporting · call centre

Supervision

Custom supervision screens embedded in the Hermès 360 telephony platform — inbound/outbound call activity and SMS volumes per campaign, read straight from versioned SQL views.

JavaScriptjQueryEChartsTabulatorT-SQL viewsSQL Serverhermes_sql bridgevendored librariesone CSS + one JS fileHermès 360 embed
production
10
multi-agent · markets

Trading Bot

Multi-agent earnings bot — reads the earnings calendar and SEC filings, scores candidates before the open, then decides buy / hold / skip. Personal lab, not financial advice.

Python 3.12SQLAlchemy 2AlembicPostgreSQLSupabasepydantic-settingsAPScheduler6 agentsIBKR · ib_insyncSEC EDGARyfinanceOpenAI-compatible LLMtenacitystructlogpytestOpenClawDocker
personal lab
view all 16 repositories on GitHub
/03what this means for you

The pipeline is the portfolio.

If you're hiring for a senior engineering role with AI, automation, or platform components — these are the patterns I'd bring to your team on day one.

I think in systems, not tickets.

The pipeline above wasn't a ticket — I saw a workflow problem, designed end-to-end, shipped. Same approach to product work.

"Production over prototype" isn't a slogan — it's how I scope.

AI is a tool, not the product.

Shipping LLM features since GPT-3.5. I know where AI breaks (eval, hallucination, cost, latency) and how to ship around it. Claude is one piece — deterministic code catches its mistakes.

"Senior" with AI means knowing when not to use it.

I run my own infra.

VPS, Docker, n8n, Postgres, Firecrawl, Telegram bots — self-hosted and hardened (OWASP, firewalls, secrets), on hardware I pay for. I'd rather understand a system than rent it.

If I can't run it locally, I don't trust it in prod.

I write French & English.

Dijon, FR — open to remote / hybrid across FR & EU. Specs, reviews, customer copy in both.

Fluent FR · senior-level EN · native UA/RU. C++ in CSS.

/04ideas & writing

Things I keep coming back to.

Groundbreaking papers, essays, and moments in AI, programming & systems — each card opens the whole article. Swipe / drag the stack.

Rotating — drag to take control
/05the stack

Tools I reach for daily.

Opinionated. Self-hosted by default. Boring where it needs to be, sharp where it matters.

right: my stack, locked · left: yours — double-click to pin a tool I'm missing
/06agent system

One orchestrator, five teams. Go ahead — poke them.

This is the system I run every day: ai-agents-config — five agent profiles, each adapted to three harnesses (Claude Code, Codex, OpenCode), fronted by Manager — my Hermes Agent AI — on my own VPS. Pick a team, hand it a real task, read the answer right here. No sign-up, no sales call.

  1. 01youone message — Telegram or CLI
  2. 02Managerreads the task, picks the team, the model and the harness
  3. 03profileone team wakes up — dev, seo, marketing, security, design
  4. 04agentsplan → build → hand off through a shared scratchpad
  5. 05gatea reviewer scores it and a verifier boots it — or it loops
  6. 06youworking software, with the evidence attached
live sandbox idle
hermes · orchestrator

Hermes

The wire between me and every team. Give it a task — it tells you who it would wake up and why.

Real model, no tools — it explains what it would do. My real agents run on my VPS with file, shell and DB access.

Want this kind of system thinking
on your team?

I'm open to AI engineering and senior full-stack roles — fully remote across EU, or hybrid in France. The CV is one click away — though honestly, this page is most of it.