Skip to content

DraftLint System Family

Status: Canonical identity and architecture map Last verified: 2026-07-16 Naming rule: DraftLint alone means this family. Use a complete variant name for a repository, service, model path, or product feature.

Why This Page Exists

DraftLint has been used for several related drawing-analysis experiments and deployments. They share history and concepts, but they do not run the same workflow and are not interchangeable.

This page is the authority for answering:

  • which DraftLint variant is being discussed,
  • where its code lives locally and remotely,
  • which branch and deployment it uses,
  • whether it performs full validation or only YOLO inference,
  • and whether it is current or historical.

Variant Registry

Canonical name Responsibility Current status Runtime
DraftLint Upstream Sreekar's original full drawing-validation codebase Active upstream Local/Docker; no Adeel-hosted production service
DraftLint Railway Validator Railway-compatible fork with gateway detection, LiteLLM, and workflow traces Active integration fork Railway Testing Service
DraftLint Detector Gateway Stable authenticated API in front of IONOS/Nuclio YOLO functions Active infrastructure IONOS
RapidDraft Drawing Analysis Customer-facing upload, detector selection, review, and findings workflow Active product Railway .ai and .io projects
DraftLint Model Training CVAT labeling, YOLO training, model promotion, Nuclio deployment, and verification Active operations Windows workspaces, cloud GPU training, IONOS CVAT/Nuclio
DraftLint AMD/Roboflow Lab Earlier Windows AMD, Roboflow, OCR, and DirectML experiment Historical Windows workstation

Source And Checkout Registry

Local paths identify the working copy on a specific machine. GitHub links identify the shared remote source. A second checkout of the same remote is listed as a working copy, not as another system.

System or checkout Local/server path Remote source Active identity Notes
DraftLint Upstream /Users/adeelyj/code/local ai server setup/DraftLint sreekarrs/DraftLint main, aaa9947 Canonical upstream checkout
DraftLint Railway Validator /Users/adeelyj/code/local ai server setup/Draftlint-forked-railway-adapter adeelyj/Draftlint-forked-railway-adapter railway-adapter, a0f3f10 GitHub default remains main; deployment code is on railway-adapter
Duplicate upstream working copy /Users/adeelyj/code/sreekar-draftlink-BOM-frontend/draftlint-source sreekarrs/DraftLint main, same aaa9947 snapshot Not a separate variant
RapidDraft primary working copy /Users/adeelyj/code/rapiddraft/45_co2/rapiddraft_utumpitch adeelyj/rapiddraft_utumpitch Local branch theegarten-rapiddraft-agent-event-stream-occ, commit b8ba988 Local checkout was dirty when verified; GitHub default is master
RapidDraft deployed source snapshot Same GitHub repository adeelyj/rapiddraft_utumpitch codex/collaboration-utility-tray, 45af414 Railway CLI deployment message for both .ai and .io on 2026-07-16
RapidDraft BOM UI working copy /Users/adeelyj/code/sreekar-draftlink-BOM-frontend/rapiddraft-bom-ui adeelyj/rapiddraft_utumpitch solidworks-agent-loop-slick-ui, 310e01d Additional checkout, not a separate product repository
RapidDraft colleague working copy /Users/adeelyj/code/sreekar-draftlink-BOM-frontend/rapiddraft-source adeelyj/rapiddraft_utumpitch theegarten-rapiddraft-agent-bom-occ, fda5cda Additional checkout
DraftLint Detector Gateway IONOS /opt/cvat-inference No Git remote recorded Operational server directory Gateway and repair scripts are present, but the directory is not a Git worktree
DraftLint Model Training Windows D:\02_Code\49_yolotraining_firstdataset No Git remote recorded Active training workspace Main CVAT + YOLO operations source
CVAT/Roboflow workspace Windows D:\02_Code\50_CVAT_RoboFlow No Git remote recorded Parallel experiment workspace Keep isolated from the main YOLO workspace
DraftLint AMD/Roboflow Lab Windows D:\02_Code\41_Training_drawing_roboflow_windowscode No Git remote recorded Historical workspace Preserved for runtime research and provenance
Product wiki /Users/adeelyj/code/rapiddraft wiki/wiki-rapiddraft adeelyj/wiki-rapiddraft GitHub default main Hosted documentation source

Architecture

flowchart LR
    UI["RapidDraft Drawing Analysis"]
    RV["DraftLint Railway Validator"]
    DG["DraftLint Detector Gateway"]
    CVAT["CVAT and Nuclio models"]
    LLM["LiteLLM and local AI server"]
    UP["DraftLint Upstream"]
    TRAIN["DraftLint Model Training"]

    UP -->|"upstream code"| RV
    UI -->|"full validation when configured"| RV
    UI -->|"current direct detector path"| DG
    RV -->|"YOLO inference"| DG
    RV -->|"optional LLM validation"| LLM
    DG -->|"model invocation"| CVAT
    TRAIN -->|"deploy and refresh"| CVAT

There are two application paths:

  1. Direct detector path: RapidDraft calls the DraftLint Detector Gateway and turns YOLO predictions into the product's review experience.
  2. Full validation path: a client submits a drawing to the DraftLint Railway Validator, which performs loading, preprocessing, gateway detection, parsing, OCR, optional LLM validation, rule assembly, and reporting.

Calling the gateway does not run the full validator. Likewise, the Railway Validator does not host the production YOLO models inside Railway.

Verified Deployment Snapshot

Deployment Railway project Services relevant here Database/storage boundary Verified behavior
app.rapiddraft.ai Deployed Service rapiddraft_utumpitch, rapiddraft-postgres One Postgres service plus /data volume 12 detectors; default 26
app.rapiddraft.io Testing Service rapiddraft_utumpitch, two separately named Postgres services, DraftLint Railway Validator App /data volume; two database services require explicit DATABASE_URL ownership 12 detectors; default 26
DraftLint Railway Validator Testing Service DraftLint Fork with Local AI No Railway volume attached when verified; uploads are cleaned after jobs and results/debug artifacts are ephemeral Health endpoint returned version 1.0.0

Both RapidDraft deployments were CLI-deployed from the same 45af414 source snapshot. They are separate Railway projects with separate database and volume resources. Never copy a database URL, Stytch secret, signing secret, or volume reference from one project into the other without an explicit migration plan.

Which Page To Use

Open Questions

  1. Should the active railway-adapter branch become the GitHub default for the Railway fork, or should it merge into main?
  2. Should /opt/cvat-inference be moved into a private infrastructure repository so gateway changes have reviewable Git provenance?
  3. Should the Windows model-training workspaces receive private remotes, or remain local operational workspaces with backed-up manifests?

Sources

  • Live Git and GitHub repository metadata verified 2026-07-16
  • Railway CLI project, service, deployment, domain, volume, and variable-name metadata verified 2026-07-16
  • Live RapidDraft detector registries and DraftLint health endpoint verified 2026-07-16
  • IONOS /opt/cvat-inference/scripts/verify-yolo-stack.sh result verified 2026-07-16
  • Central playbook 04_playbooks/DRAWING_MODEL_TRAINING_AND_CVAT_MODEL_SERVING.md
  • Central handover 01_tracks/techstack/handover/260603_draftlint-yolo-gateway-railway-stabilization.md