Freelancer profile translated to English.
Back to original languageAbout Olivier
I design and integrate artificial intelligence solutions into production environments:
- AI-driven autonomous websites,
- Business process automation,
- Connecting AI APIs to existing systems (CRM, ERP, e-commerce).
Concrete example: a fully autonomous WordPress site where AI reads specialized press, generates a summary, produces an image, and publishes the article — without human intervention, with long-term SEO and editorial monitoring.
My approach: start with a real business need (not an AI gadget), build a reliable architecture (data validation, error handling, traceability), and deliver a system that runs autonomously in production.
15 years of development (PHP, JavaScript/TypeScript, Python) allow me to integrate AI into existing stacks rather than proposing isolated solutions.
French
Native or bilingual
English
Fluent
German
Basic
Remote only
Primarily works remotely
Experience
- Marque BlancheLocal Market Viability Scoring Tool — Cross-referencing Public Data and AI SynthesisTECHDecember 2025 - July 2026 (7 months)Châteaurenard, FranceSolo design and development of a tool that answers a simple but poorly served question: is a local market saturated or still open for a given business? Instead of relying on indirect indicators (visible competition, reviews, approximate Google searches), the tool cross-references structured public data to produce a defensible score.How it works:
- Cross-referencing two families of heterogeneous data: online demand and competition signals via the DataForSEO API (search volume, organic competition density), and local economic data via INSEE/SIRENE public APIs (existing establishments, demographic dynamics).
- Viability score out of 100, broken down into four weighted factors (demand, market accessibility, local dynamics, competitive pressure).
- Raw data is sent to the OpenAI API to generate a clear language summary — the score alone is not enough to make the result usable for a non-specialist user.
Technical architecture:- Next.js, Vercel deployment, serverless PostgreSQL on Neon — infrastructure chosen so that costs remain proportional to actual usage rather than reserved capacity.
- Authentication via NextAuth (GitHub login).
Main challenge: reliability and cross-referencing of sources with different update frequencies and granularities (web search data vs. administrative data), without one masking the blind spots of the other.Result: functional tool built alone, from scratch, with reproducible score calculation. The initial weighting of factors is documented as a working hypothesis, to be confronted and adjusted against real cases.Key takeaway: AI synthesis is only valuable if it's based on verifiable and cross-referenced data — AI presents and explains a result, it doesn't replace the reliability of the upstream data. - Oramus SASOramus — Applied AI Lab: Design, Development, and Management in Complete AutonomySOFTWARE PUBLISHINGJune 2026 - June 2026Châteaurenard, FranceSolo design and development of an Applied AI Lab for SMEs: an online diagnosis that analyzes a website and identifies 3 to 5 concrete AI use cases in 30 seconds, with authentication, user quota, PDF report, and GDPR compliance. A complete product in production, not a prototype.Architecture:
- Astro 5 (SSR + static), Cloudflare Pages, Cloudflare D1 (serverless SQLite).
- Anthropic Claude API in streaming (Server-Sent Events): generation of the diagnosis in structured JSON, validated and parsed server-side.
- 5-step pipeline: HTML fetch and compression (200 KB → 4-8 KB useful signal), technical stack detection, competitive enrichment via SEO API, streaming AI call, persistence with automatic 30-day retention.
- Operating cost: €10 to €20/month (infra + DB), API usage per request.
Working method with AI — three distinct and deliberately separated channels:- Design and architectural arbitration: exploratory dialogue, decision always human.
- Implementation: development agent under continuous supervision (systematic review of diffs, validation or correction at each step).
- Production: AI as a component of the product delivered to the end-user, with requirements for reliability and controlled costs.
Technical challenges resolved: premature Anthropic streaming closure bug on Cloudflare Workers (incorrect SDK pattern), behavior difference between local and production environments requiring systematic validation in production, dependency management with strict security policy (daily published packages).Key takeaway: added value in an AI-driven project comes not from the tool but from the ability to maintain the architecture and segment decision levels — a poorly managed project drifts, a well-managed project saves considerable time without sacrificing rigor. - Oramus SASCryptoCycleMonitor — Bitcoin Analysis Dashboard with AI-Generated SummariesPRIVATE EQUITYJune 2026 - June 2026Châteaurenard, FranceDesign and development of a public dashboard aggregating 7 Bitcoin on-chain and technical indicators (MVRV Z-Score, NUPL, SOPR, RSI, Fear & Greed, Funding Rate, distance to ATH) into a single composite score, accompanied by a weekly natural language summary generated by AI.Stack and architecture:
- Next.js 14 (App Router), PostgreSQL on Neon, Vercel hosting with cron jobs for daily data collection.
- Integration of several third-party APIs (BGeometrics, CoinGecko, Binance, alternative.me) with strict quota management (15 req/day on the main source) and caching strategy to stay within free limits.
- Integration of the Anthropic API (Claude Haiku): structured injection of the week's indicators into a prompt, generation of an editorial summary of 200-300 words contextualizing the figures.
Assumed technical choices:- Deliberate selection of the most economical AI model suited to the task (Haiku rather than Sonnet/Opus): narrative formatting of already structured data does not justify a more expensive model — final AI operating cost: less than $0.01/month.
- Composite score with defined weighting and normalization by documented empirical thresholds, rather than an opaque system.
- Editorial choice to never generate buy/sell signals: methodological discipline rather than impressive but unfounded functionality.
Result: tool in production, total operating cost less than €1/month (infrastructure + AI), public documentation of the system's methodological limitations.Key takeaway: AI integration in production is judged on its economic relevance and methodological rigor, not on the power of the chosen model — use the most suitable model for the actual need rather than the most impressive one.
Reviews
Elsa
En trois clics
Reviewed on 10/11/2022
Great collaboration! Fast and efficient, thank you for everything :-)
Recommendations
+3
These freelancer profiles also match your criteria
Agatha Frydrych
Backend Java Software Engineer
4.7
(3)
2
Baptiste Duhen
Fullstack developer
4.6
(4)
5
Amed Hamou
Senior Lead Developer
4
(2)
7
Audrey Champion
Web developer
4.3
(3)
4
Education
- Consular Diploma in Management and CommerceKedge2002Formation en gestion d'entreprise ( marketing, communication, comptabilité, droit,commerce, gestion analytique)