Skip to content

Curriculum Vitæ

Boris Dupoux — CV: applied AI, computer vision, remote sensing and industrial process design. 14+ years across industry, R&D and entrepreneurship.

↓ Download full CV (PDF)

Boris Dupoux — Director & Co-Founder, Capture (Process Mapping & Automation). French · born 1986 · Languages: French (native), English (fluent), Mandarin (basic), German (basic).

Research interests & expertise

My work sits at the intersection of industrial engineering, applied AI and operational process design. Over 14+ years I have developed deep practical expertise in:

  • Process mapping, analysis and optimisation in complex multi-site environments
  • Computer vision for industrial inspection: crack detection in civil structures, object detection and tracking in fixed-camera and drone video
  • Remote sensing & predictive modelling: satellite-based algae detection and stranding prediction, drone multispectral crop assessment, day/night pollutant detection
  • 3D vision & geopositioning: monocular depth, ray-cast geolocation of objects in video and inter-object distances; photogrammetry and Gaussian Splatting from handheld video
  • Speech & local AI agents: transcription and diarisation pipelines, hardened local LLM agents, knowledge graphs with provenance and human validation
  • Document automation & intelligent data processing (OCR, NLP, cross-document reconciliation)
  • ERP integration, lifecycle data management and digital transformation of legacy workflows
  • Product development & industrialisation (design-to-production, international supply chains)

I am particularly interested in applied research that bridges AI/automation and real-world industrial operations — translating laboratory methods into deployable, measurable solutions.

Education

  • 2006–2009 — Diplôme d’Ingénieur (M.Eng. equivalent) — École Centrale Marseille, France Project Management & Industrial Engineering; Entrepreneurship & Business Management track.
  • 2003–2006 — Classe Préparatoire aux Grandes Écoles (CPGE PC*) — Lycée Carnot, Dijon.

Professional experience

  • 2024–present — Director & Co-Founder, Capture (Martinique) — process mapping & document-automation company serving 15 client groups. Proprietary Python/LLM tooling that halved delivery time; measured x2–x10 ROI across transit, wholesale, retail and accounting.
  • 2024–2026 — Co-Founder, Carige (Martinique) — drone & satellite technical data acquisition and processing for agriculture, environment and infrastructure. Computer-vision pipelines (crack detection, video geolocation), Sargassum detection, monocular depth, day/night pollutant detection.
  • 2024 — After-Sales Director (49 reports), Groupe Citadelle — €300M automotive group, 800+ employees, 15 brands. +15% throughput, +5pts retention; KPI dashboards.
  • 2022–2024 — Customer Experience Director (27 reports), Groupe Citadelle — built the CX function from scratch (call centre, CRM, omnichannel, e-commerce UX/UI).
  • 2021–2022 — Head of Business & Technology Development, Groupe Citadelle.
  • 2013–2017 — Founder & CEO, Vibe — decorative lighting manufacture; French Tech Innovation Grant (BPI France, 2016).
  • 2009–2012 — Product Manager, R&D & International Sales, Smart & Green — co-inventor of a wireless device-pairing method (WO2014016503, granted in France as FR2993731): a control module pairs luminaires that are out of reach once installed — floating in a pool, hung high in trees — without touching them, by broadcasting a low-power universal configuration signal to which each unpaired device answers with its unique identifier. Wireless LED startup; product line built from zero, industrialisation in China, sales in 13 countries; second patent on a modular LED luminaire (EP2532944).

Teaching

  • 2021–2022 — Lecturer, ICEA Catholic Institute (HR Management, Strategy & Leadership, Digital Culture).
  • 2018–2021 — Workshop Facilitator & Coach, CCI Martinique — Village by CA accelerator.
  • 2014–2015 — Tutor, Internat de la Réussite — academic mentoring for underserved students.

Patents

  1. WO2014016503 (PCT application; granted in France as FR2993731, 2015) — Device pairing: contact-free pairing of hard-to-reach wireless devices with a control module (2014).
  2. EU Patent EP2532944 — Modular LED Lighting Module and Luminaire (2012).

Applied R&D projects

Thirteen projects combining data acquisition, algorithm development and real-world operation — see Research →:

  • Vision & 3D: geolocated video analysis and inter-object distances; handheld video to photogrammetry and Gaussian Splatting; sub-millimetre crack detection.
  • Remote sensing: Sargassum detection and stranding prediction; day/night pollutant detection.
  • Documents, speech & automation: a library of explainable document-processing atoms; multitrack recordings to an editable AAF timeline; deterministic process automation without LLM; field identification of recall-affected vehicles.
  • Data, knowledge & agents: urban flows of Fort-de-France from open data; mission knowledge graphs queried through MCP; a fully local, hardened AI agent.
  • Signal processing: Codex — high-density archiving of media on paper.

Technical skills

  • LLMs & agents: local-first LLM integration — Ollama, llama.cpp and MLX-LM running Qwen3 (8B, 14B, 30B-A3B), Qwen3-VL, Qwen2.5-Omni, gpt-oss and Mistral — plus the Mistral, Anthropic (Claude) and OpenAI APIs; a hardened, fully local personal agent built on Hermes Agent with custom MCP servers (read-only sources, no shell, remote access over VPN + TOTP); RAG with hybrid search (SQLite full-text + vectors, bge-m3 embeddings, bge-reranker-v2-m3); schema-validated JSON extraction with verbatim-citation checks.
  • Computer vision: detection, segmentation and tracking (YOLO11 detect/seg, RT-DETR, SAM 2, SegFormer, ByteTrack, RAFT optical flow), monocular depth (Depth Anything V2, Depth Pro), multimodal image search (SigLIP), thin-structure detection (Frangi), licence-plate reading, video geolocation — OpenCV, Ultralytics, PyTorch, ONNX Runtime.
  • Speech & audio: transcription and diarisation pipelines (Whisper large-v3 / turbo, faster-whisper, WhisperX with wav2vec2 alignment, pyannote 3.1, Silero VAD), Whisper hallucination filtering, multi-microphone alignment and AAF export for film editing, voice cloning (XTTS v2).
  • Documents & OCR: hybrid native-text/OCR PDF pipelines (PaddleOCR, Tesseract, EasyOCR, Qwen3-VL), layout analysis, cross-document reconciliation and explainable validation; an in-house atoms library of 450+ tested Python modules assembled into YAML recipes (~3,000 tests).
  • Compute architecture: one codebase across NVIDIA CUDA (RTX 5090 workstation) and Apple Silicon (PyTorch MPS, MLX, Metal llama.cpp) with automatic cuda → mps → cpu selection; benchmarked model choices; GPU locking and scheduled jobs.
  • Data, web & GIS: Python, SQL (MySQL/MariaDB, SQLite, PostGIS), FastAPI, PHP, React, Hugo, Splink record linkage, Google Apps Script, ERP integration; GIS and photogrammetry (QGIS, Agisoft Metashape), satellite imagery analysis.
  • Infrastructure: Linux/macOS/Windows servers, Docker, VMware/ESXi, NetBird mesh VPN, shared-hosting deployment (Hostinger).
  • Hardware: drone operations (multispectral, thermal/IR, LiDAR, high-res RGB), photogrammetry, industrial prototyping.
  • Process & management: process mapping, BPM, KPI design, CRM deployment, change management, Lean.

International experience

3 years based in China (Shenzhen/Guangzhou); commercial operations across 13 countries; multi-territory management (5 French overseas departments + Iceland); professional collaborations with Scandinavian design studios (Georg Jensen, Danish design studios).