Pavlo Puzikov

Profile

Innovation researcher working across product research, applied AI, and go-to-market. I run technology and trend research, prototype the systems that come out of it - immersive 3D, autonomous agents, computational valuation - and brief stakeholders on where the next bets are.

Twelve production applications shipped end-to-end. Two brands represented on-stand at four international tech expos. Single-operator using agentic AI workflows.

Core competencies

• Artificial intelligence and machine learning

  Fine-tuning, retrieval, multi-agent systems, computer vision. Large language model fine-tuning (LoRA, QLoRA), Retrieval-augmented generation, Multi-agent orchestration, Agentic workflows, Prompt engineering, Computer vision, 3D Gaussian Splatting.

• Engineering

  Production code authored alongside SOTA coding agents (Claude Code, Codex, Antigravity). Strongest at architecture, integration, and shipping. TypeScript, Python, Next.js, React, Node.js, FastAPI, PyTorch, PostgreSQL, Prisma, Tailwind, Vercel, AWS, Docker, Linux.

• Product & Research

  Trend scouting to value-prop validation to roadmap framing. Technology and trend scouting, Immersive-tech and AI prototyping, Value-proposition testing, User research, Roadmap framing, Competitive teardowns.

• Marketing & Business Development

  International technology expos, partnerships, and demand generation across four languages. GITEX, LEAP on-stand business development, Partnership development, Brand activation, Creative direction (graphics, copy), Search and answer-engine optimisation, Multilingual content, LinkedIn and content strategy, Demand generation.

• Data and intelligence

  Benchmarking, sizing, correlation, geospatial risk, business intelligence. Traffic & performance benchmarking, Market sizing, Correlation analysis, Audience research, Web-data extraction, Geospatial risk modelling, Apache Superset, PostHog.

• 3D & Visualisation

  Real-time 3D, photogrammetry, AI-assisted media for storytelling. 3D Gaussian Splatting, Photogrammetry, LiDAR scanning, Cesium, Three.js, Blender, Real-time 3D, AI-assisted media.

Tools

• AI Coding Agents.Claude Code, Google Antigravity, OpenAI Codex, Cursor, GitHub Copilot.
• Generative Media.Midjourney, Sora, Runway, Veo, Flux, Stable Diffusion, AI voice & music.
• 3D & Spatial.3DGS capture & editing, LiDAR / photogrammetry, Cesium, Three.js, Blender.
• Web Design & Build.HTML/CSS, Figma, Framer, Webflow, Wix, Squarespace.
• Productivity.Notion, Linear, Twenty CRM, HubSpot, LinkedIn Sales Navigator, Google Workspace.
• Analytics and business intelligence.Apache Superset, PostHog, Google Analytics, Similarweb, Semrush, Ahrefs.

Experience

1. Product Innovation Researcher · BARNES Dubai

   Jun 2025 - Present

   Dubai, UAE

   Lead the technology and innovation research function. Single-operator product, ML, and applied research across ten production applications. BARNES is the world's largest international luxury real-estate network (60+ countries).

   • Built and shipped Atlas - entity-centric valuation platform with hex-grid geospatial risk modelling and six-month price forecasts.
   • Fine-tuned Barnes AI on ~9,600 instruction pairs and deployed via Ollama; wrapped in a multi-agent research system that turns days of analyst work into on-demand briefs.
   • Productionised the 3DGS research pipeline (scanned interiors, AI-generated scenes for unbuilt properties, city-scale capture). Submitted to SIGGRAPH Asia 2026.
   • Stood up the Dubai Digital Twin; first georeferenced building (Onyx Tower) is live in property-search flow.
   • Built the Innovation Portal - executive cockpit with project status, valuation analytics, competitive signal, multilingual RBAC, embedded Apache Superset business intelligence.
   • Shipped Splat Viewer (web-native 3DGS with mobile tap-to-move), Smart Deal Global (public marketplace), and multilingual marketing surfaces - ten production web apps in twelve months.

2. Marketing & Partnerships Lead · Living Homes

   Jun 2025 - Present

   Dubai, UAE · promoted from Market Research Specialist

   Own marketing, partnerships, and digital presence for a luxury smart-home developer. Co-drive global expansion alongside BARNES Dubai.

   • Run the brand's official LinkedIn and the wider digital-channel mix - drives qualified inbound from regional partners and end clients.
   • Represented Living Homes & BARNES on-stand at GITEX Global 2024/2025 (Dubai), LEAP 2025 (Riyadh), GITEX Asia 2025 (Singapore). Helped expand both brands into Saudi Arabia and Southeast Asia.
   • Continuous competitive research across traffic, positioning, pricing, features - outputs feed product, sales, and partnership decisions directly.
   • Track the broader innovation and technology landscape (AI, immersive tech, PropTech) and brief stakeholders on opportunities for the product.
   • Coordinate regional partner relations across the Gulf and EU; produce multilingual marketing materials.

3. Market Research Specialist · Living Homes

   Sep 2024 - Jun 2025

   Sofia, Bulgaria

   Studied emerging AI and smart-home applications across EU and UAE real-estate markets.

   • Built a custom multi-agent research system that automates competitive-website analysis, structured extraction, and ranked-insight reporting.
   • Analysed positioning, traffic, and feature sets across 50+ smart-home and PropTech vendors; surfaced differentiation gaps and partnership opportunities.
   • Tracked advancements in applied AI, computer vision, and spatial computing; distilled implications into executive strategy briefs.
   • Produced insight artefacts (market maps, opportunity briefs, competitor teardowns) used to inform expansion and partnership decisions.

4. Web Design & Development Intern · W Korea

   Jun 2024 - Aug 2024

   Seoul, South Korea (remote)

   • Contributed to web design and front-end development workstreams within a brand-led editorial environment; produced layout, motion, and asset work for a flagship fashion-publishing title.

5. Marketing Intern · Hyundai Motors New Zealand (Inspire Marketing)

   Nov 2022 - Feb 2023

   Auckland, New Zealand

   • Delivered campaign work across Hyundai Motors NZ, Isuzu Utes NZ, and Renault NZ spanning advertising, web, social, market research, email, and copywriting.

6. Customer Vehicle Representative · Hertz New Zealand

   Jul 2022 - Nov 2022

   Auckland, New Zealand

   • Customer-journey engagement and operations work across the rental lifecycle.

Education

• Bachelor of Commerce - Marketing and Management

  The University of Auckland

  Aug 2020 - Dec 2023

• Secondary Diploma

  Scots College, Wellington, New Zealand

  Feb 2014 - Nov 2019

Speaking and events

• 2025GITEX Global 2025Dubai, UAEOn-stand business development · Living Homes & BARNES Dubai
• 2025GITEX Asia 2025SingaporeRegional expansion · Living Homes & BARNES Dubai
• 2025LEAP 2025Riyadh, Saudi ArabiaSaudi market entry · Living Homes & BARNES Dubai
• 2024GITEX Global 2024Dubai, UAEOn-stand business development · Living Homes

Selected research and output

• Where the field is in 2026

  Zillow shipped 3DGS via SkyTours in 2026; Apartments.com responded with Matterport 3D Exteriors. Khronos ratified KHR_gaussian_splatting for glTF 2.0 in February. World Labs released Spark 2.0 (streamable 100M+ splat LOD). Apple released SHARP (single-image 3DGS). Onyx Tower has been live in BARNES property-search since before any of this.

• SIGGRAPH Asia 2026 - Tech Communications (submitted)

  Applied 3D Gaussian Splatting in luxury real-estate visualisation. Pipeline spans interior scans, AI-generated scenes for unbuilt properties, and city-scale capture.

• HERO 3DGS Pipeline

  Unified processing pipeline consolidating 21 sub-pipelines into 9 stage adapters. Two paths: city-scale (single-image Lyra from Street View) and property scan (XGRIDS PortalCam .xbin). Two viewer modes: city flythrough and shareable property link.

• LangSplat — natural-language scene queries

  CLIP-embedded gaussians let users search a 3D scene by text ("show me the kitchen") and fly the camera there. CVPR 2024 method, wired against the Splat Viewer.

• Furniture Removal — virtual de-staging

  YOLO detect → SAM segment → Stable Diffusion inpaint → MiDaS depth. Strips a scanned interior down to walls and floor so a buyer can re-stage virtually.

• Seen2Scene — flow-matching hole completion

  Partial scan + TSDF + visibility mask → flow-matching → complete geometry. Fills the holes that LiDAR can't see (under sofas, around tight corners).

• 4D Gaussians — temporal capture

  Time-varying 3D scene capture for construction progress. Monthly photogrammetry passes become smooth 3D timelapses; the same building, rendered at any timestamp.

• Reflection-Aware 5DGS

  Extension of the 3DGS representation to handle reflective and specular surfaces (glass, polished marble, mirrors) — the classic failure mode of vanilla 3DGS in luxury interiors.

• Floor Plan Generator + Property Semantic Memory

  Open3D RANSAC walls + CLIP furniture classification produces architectural floor plans from raw .lci scans in under a second. Paired with YOLO + Mask2Former scene-graph memory so the system can answer "how many bedrooms?" without LLM guessing.

• GaussForge — lossless format converter

  High-performance lossless conversion between PLY, SPZ, SPLAT, and KSPLAT. Fed into the Splat Viewer (needs .splat), gaussian-compression (needs .ply), and web delivery (needs .ksplat).

• Internal publications

  Product-innovation reports, competitive teardowns, market briefs, valuation methodology notes, and open-source tooling around 3DGS and multi-agent orchestration. Available on request.