Sébastien Burel — haruni.net
Computer engineer graduated from EPITA (1993), I've been building innovative digital products for over 20 years through haruni.net. Today, my primary focus is generative AI: orchestrating multi-model LLM systems (OpenAI, Gemini, Llama, Anthropic, Mistral, AWS Bedrock) to build intelligent, high-performance applications. Based in Rennes, France, I work remotely for clients across France and internationally.
View my LinkedIn profile →Services
Generative AI
Design and development of LLM-based applications: RAG, autonomous agents, fine-tuning, multi-model orchestration (OpenAI, Gemini, Mistral, AWS Bedrock, Llama).
AWS Cloud
Architecture, deployment and optimization of cloud applications on AWS: EC2, Lambda, DynamoDB, S3, Bedrock, API Gateway.
Mobile & Web
Development of iOS, Android and web applications: React, Node.js, HTMX — from MVP to production.
Architecture & CTO as a Service
Technical guidance for startups and SMBs: architecture design, code review, technical leadership, team upskilling.
By the numbers
Projects
Frigo Magic
Anti-food-waste recipe app based on ingredients available at home. 3 million downloads, 10 million meals saved per year.
CraftJS
Mac desktop application leveraging large language models: OpenAI, Gemini, Llama, Anthropic, Mistral via AWS Bedrock. Voice processing with Whisper and PyTorch.
bymuriel.fr
Artisan candle and wax melt shop. AI image generation pipeline with ComfyUI + Flux2 on DGX Spark, custom Node.js backoffice and Shopify sync.
CornerNews
Real-time photo-sharing iOS app for breaking news. Users become citizen reporters by posting geolocated content.
Latest articles
Generating product visuals for a Shopify store with ComfyUI, Flux2 and a DGX Spark
AI image generation pipeline for an artisan shop: 132 professional visuals produced automatically from a custom Node.js backoffice.
Predicting Vilaine River Floods with Deep Learning
How I built an LSTM + Attention model to predict water levels across 11 stations.
Image Generation on the DGX Spark: 11 Models Compared
I benchmarked 11 image generation models on the NVIDIA DGX Spark — FLUX, Stable Diffusion, Qwen, Z-Image — same prompt, same seed, same conditions.