A.I. Research Project | The Universe
The Next Horizon: Local LLMs & Autonomous Built-Environment Platforms
-
As HTDSTUDIO surpasses three years of dedicated AI research and implementation, our focus expands from external model orchestration ("prompting cloud engines") toward locally-hosted open-weight architectures (e.g., Qwen 27B series). By bringing high-capacity model inference directly onto our local hardware and edge environments, we are scaling our internal capabilities, protecting project intellectual property, and building high-speed, localized spatial tools.
Key Focus Areas for 2026 & Beyond:
1. Sovereign & Local Model Architecture
- Data Privacy & IP Security: Local inference eliminates external API dependencies for sensitive client briefs, site feasibility studies, financial modeling, and proprietary design logic.
- Deep Context & Low Latency: Running optimized ~27B parameter models locally allows near-instant agentic loops, continuous long-context reasoning over local building codes, and rapid iterations during active design workshops.
2. Autonomous Agentic Systems for Architecture & Real Estate
- Scaling localized multi-agent workflows that run continuously in the background to handle:
- Feasibility & Zoning: Parsing Swedish (Boverket) and Portuguese regulatory codes alongside local market comparables.
- Circular Economy Sourcing: Autonomous cross-referencing between CAD/BIM material schedules and local reclamation databases.
- Vibe Coding & Automated Tooling: Accelerating custom iOS/Swift utility apps and internal plugin development directly within Xcode and studio pipelines.
3. Boutique APIs & EU/US Compliance Platforms
- Integrating local model logic into tailored API microservices for partners and networks.
- Delivering automated tracking for GHG emissions, circular material lifecycles, and EU taxonomy/sustainability compliance across supply chains without exposing client datasets.
4. Edge & On-Site Mobile Intelligence
- Combining Apple Silicon local compute power with iOS app development (TestFlight/Xcode) to deploy compact local LLM agents into field research, site surveys, and real-time design reviews.
-
Client: Experimental | Location: The Universe | Total Floor Area: ♾️ ft2/m2 | Budget: ♾️ | Status: 2023-Ongoing. Software: Apple Simulator / Apple TestFlight / Apple Xcode / ChatGPT5.2 / Claude Sonnet 4.5/ Figma / Fotor Ai extension / Gemini / Google AI Studio / Google Firebase / MidJourney V7.0 / Nano Banana Pro / Notebook LM / Repl.it / Supabase / Veo2 / WAN AI / Windsurf | Hardware: Apple MacBook Pro M2Max 38-core / Apple iMac 2020 27” 80GB / 10TB/oftware additions: Qwen 27B (Local Inference) / Ollama / LM Studio / vLLM / LocalAI / PyTorch / Apple MLX Framework / Xcode / Supabase / Firebase / Claude Sonnet / ChatGPT
Hardware / Infrastructure updates: Highlight high-memory local inference nodes (e.g., Apple Silicon Unified Memory Workstations / Local GPU Clusters for 27B+ unquantized/FP16 or 4-bit/8-bit quantized runs).
-
Concept Design / Prompt Engineering / Vibe Coding: HTDSTUDIO
-
Guest Lecturer: “Negotiating with the Genie”, The Hive: "Artificial Intelligence in the Field", Capgemini Applied Innovation Exchange, Malmö, Sweden - November, 2023
(Above) Source code for creating surface tessellation with rapid prototyping, designed with the assistance of ChatGPT5.
