Skip to content

RashiD2801/NAMELab-Public

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 

Repository files navigation

NAMELab

AI-assisted design tools for architecture practice — semantic image search, knowledge extraction from past projects, and a conversational concept development engine that learns how the studio thinks.

An ongoing research and development initiative by NAME Architecture exploring how artificial intelligence can enhance the way architects search, think, and develop ideas.


Demo

Watch the demo video


Overview

Architecture practices accumulate an extraordinary body of knowledge over time: thousands of reference images, years of design conversations, audio recordings, project briefs, and the reasoning behind every design decision. Most of this knowledge is invisible — stored in formats that are difficult to search and rarely available when a new project begins.

NAMELab organises that knowledge, makes it searchable, and puts it into conversation with new work.


Tools

Studio Image Library

A semantic search and visual exploration environment for large architectural image collections.

Search using natural language: a spatial quality, a material, a mood, an architectural element. Queries are expanded into precise architectural synonyms and matched across 16 metadata fields — spatial quality, materiality, light, mood, colour tone, and more. Results are ranked by relevance and weak matches are filtered automatically.

The full library is laid out on a visual canvas using UMAP clustering, which organises images spatially by similarity so related images sit together. Search results can be isolated from the rest of the collection or explored in context. Images can be drag-selected across the canvas and downloaded in bulk, enabling quick access to pre-approved reference images for use across the studio.


Concept Development Tool

A knowledge extraction and conversational design tool that learns from a studio's completed work.

Audio recordings from design discussions are transcribed and combined with project documents. GPT-4o extracts the underlying reasoning into a structured format: the first observation on site, the question that preceded the concept, the core paradox, the concept generator, the rule system, and the material strategy. This becomes a searchable knowledge base that grows with every project.

A second layer captures how the studio thinks across all projects — core principles, diagnostic questions by project type, and the logical sequence of design enquiry. This document is updated over time as new projects reveal new patterns, always with human review before any change is written.

Before the conversation begins, three parallel analysis agents run automatically: one covering site analysis, one researching the contextual and vernacular character of the area, and one benchmarking relevant reference projects. Their findings feed into an editable briefing report that the design team can review and refine. Once the report is ready, the conversation begins with the design team already informed.

When a new project begins, the tool analyses the site context, identifies architectural patterns in the surrounding built environment, and enters into conversation with the design team. It asks questions calibrated to the brief, building on ideas already expressed. The exchange is generative: it listens, probes, and develops the thinking in directions aligned with how the studio has learned to work.


Future Development

Continuous learning loop for the image library. As the library is used, keywords and metadata can be reviewed, corrected, and expanded directly within the tool. If a search returns unexpected results, the tags can be adjusted and the index updated — so the library improves with use rather than remaining static.


Tech Stack

Layer Technologies
Backend Python 3.11, FastAPI, Uvicorn, SQLite
Frontend Pixi.js v8 (WebGL), Vanilla JS
AI / ML GPT-4o, GPT-4o-mini, Whisper, text-embedding-3-small, UMAP
Document processing pdfplumber, python-docx
Infrastructure Docker, Docker Compose, Caddy, Hetzner VPS

Roadmap

  • Studio Image Library — semantic search, UMAP clustering, isolate results, drag-to-select, bulk download
  • Concept Tool — audio transcription, document extraction, structured project JSON
  • Studio reasoning layer — living principles document, pattern detection across projects
  • RAG retrieval — embedding-based search over extracted project knowledge
  • Conversational concept development — site analysis, contextual pattern recognition, generative dialogue
  • Continuous keyword learning loop — editable metadata, in-tool corrections, index updates
  • Three analysis agents — site, context and vernacular, benchmarking (parallel)
  • Editable briefing report — browser-based, downloadable as PDF

Status

Active development. The Studio Image Library is live and in use. The Concept Development Tool extraction pipeline is complete and being run across the studio's project archive.

About

Designed and built end-to-end within NAMELab: a CLIP semantic search system across 10,000+ architectural images with a GPT-4o Vision tagging pipeline classifying each image across 15 architectural attributes, deployed for internal studio use via FastAPI and Docker.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors