diff --git a/.github/workflows/docker-publish.yml b/.github/workflows/docker-publish.yml
index 62f34846..7bfe0f9c 100644
--- a/.github/workflows/docker-publish.yml
+++ b/.github/workflows/docker-publish.yml
@@ -5,17 +5,22 @@ on:
   workflow_call:
     inputs:
       tag_suffix:
-        description: 'Custom tag suffix for the Docker image'
+        description: "Custom tag suffix for the Docker image"
         required: false
         type: string
-        default: ''
+        default: ""
       is_nightly:
         description: 'Whether this is a nightly build'
         required: false
         type: boolean
         default: false
+      skip_multiarch:
+        description: "Skip multi-architecture build for faster CI"
+        required: false
+        type: boolean
+        default: false
   push:
-    branches: [ "main" ]
+    branches: ["main"]
 
 jobs:
   build_and_push_extproc:
@@ -23,88 +28,117 @@ jobs:
     permissions:
       contents: read
       packages: write
+    strategy:
+      matrix:
+        image: [extproc, llm-katan]
+      fail-fast: false # Continue building other images if one fails
 
     steps:
-    - name: Check out the repo
-      uses: actions/checkout@v4
-
-    - name: Set up Docker Buildx
-      uses: docker/setup-buildx-action@v3
-
-    - name: Set up QEMU
-      uses: docker/setup-qemu-action@v3
-
-    - name: Log in to GitHub Container Registry
-      uses: docker/login-action@v3
-      with:
-        registry: ghcr.io
-        username: ${{ github.actor }}
-        password: ${{ secrets.GITHUB_TOKEN }}
-
-    - name: Generate date tag for nightly builds
-      id: date
-      if: inputs.is_nightly == true
-      run: echo "date_tag=$(date +'%Y%m%d')" >> $GITHUB_OUTPUT
-
-    - name: Set lowercase repository owner
-      run: echo "REPOSITORY_OWNER_LOWER=$(echo $GITHUB_REPOSITORY_OWNER | tr '[:upper:]' '[:lower:]')" >> $GITHUB_ENV
-
-    - name: Build and push extproc Docker image
-      uses: docker/build-push-action@v5
-      with:
-        context: .
-        file: ./Dockerfile.extproc
-        platforms: linux/amd64,linux/arm64
-        push: ${{ github.event_name != 'pull_request' }} # Only push on merge to main, not on PRs
-        tags: |
-          ${{ inputs.is_nightly == true && format('ghcr.io/{0}/semantic-router/extproc:nightly-{1}', env.REPOSITORY_OWNER_LOWER, steps.date.outputs.date_tag) || format('ghcr.io/{0}/semantic-router/extproc:{1}', env.REPOSITORY_OWNER_LOWER, github.sha) }}
-          ${{ inputs.is_nightly != true && format('ghcr.io/{0}/semantic-router/extproc:latest', env.REPOSITORY_OWNER_LOWER) || '' }}
-
-  build_and_push_llm_katan:
-    runs-on: ubuntu-latest
-    permissions:
-      contents: read
-      packages: write
-
-    steps:
-    - name: Check out the repo
-      uses: actions/checkout@v4
-
-    - name: Set up Docker Buildx
-      uses: docker/setup-buildx-action@v3
-
-    - name: Set up QEMU
-      uses: docker/setup-qemu-action@v3
-
-    - name: Log in to GitHub Container Registry
-      uses: docker/login-action@v3
-      with:
-        registry: ghcr.io
-        username: ${{ github.actor }}
-        password: ${{ secrets.GITHUB_TOKEN }}
-
-    - name: Generate date tag for nightly builds
-      id: date
-      if: inputs.is_nightly == true
-      run: echo "date_tag=$(date +'%Y%m%d')" >> $GITHUB_OUTPUT
-
-    - name: Set lowercase repository owner
-      run: echo "REPOSITORY_OWNER_LOWER=$(echo $GITHUB_REPOSITORY_OWNER | tr '[:upper:]' '[:lower:]')" >> $GITHUB_ENV
-
-    - name: Extract version from pyproject.toml
-      id: version
-      run: |
-        VERSION=$(grep '^version = ' e2e-tests/llm-katan/pyproject.toml | sed 's/version = "\(.*\)"/\1/')
-        echo "version=$VERSION" >> $GITHUB_OUTPUT
+      - name: Check out the repo
+        uses: actions/checkout@v4
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Set up QEMU (only for multi-arch builds)
+        if: inputs.skip_multiarch != true
+        uses: docker/setup-qemu-action@v3
+        with:
+          platforms: arm64
+
+      - name: Log in to GitHub Container Registry
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Generate date tag for nightly builds
+        id: date
+        if: inputs.is_nightly == true
+        run: echo "date_tag=$(date +'%Y%m%d')" >> $GITHUB_OUTPUT
+
+      - name: Set lowercase repository owner
+        run: echo "REPOSITORY_OWNER_LOWER=$(echo $GITHUB_REPOSITORY_OWNER | tr '[:upper:]' '[:lower:]')" >> $GITHUB_ENV
+
+      # Rust build cache for extproc - only use GitHub Actions cache for non-PR builds
+      - name: Cache Rust dependencies (extproc only)
+        if: matrix.image == 'extproc' && github.event_name != 'pull_request'
+        uses: actions/cache@v4
+        with:
+          path: |
+            ~/.cargo/bin/
+            ~/.cargo/registry/index/
+            ~/.cargo/registry/cache/
+            ~/.cargo/git/db/
+            candle-binding/target/
+          key: ${{ runner.os }}-cargo-extproc-${{ hashFiles('**/Cargo.lock', '**/Cargo.toml') }}
+          restore-keys: |
+            ${{ runner.os }}-cargo-extproc-
+
+      # Set build context and dockerfile based on matrix
+      - name: Set build parameters
+        id: build-params
+        run: |
+          if [ "${{ matrix.image }}" = "extproc" ]; then
+            echo "context=." >> $GITHUB_OUTPUT
+            echo "dockerfile=./Dockerfile.extproc" >> $GITHUB_OUTPUT
+            echo "platforms=${{ inputs.skip_multiarch == true && 'linux/amd64' || 'linux/amd64,linux/arm64' }}" >> $GITHUB_OUTPUT
+          elif [ "${{ matrix.image }}" = "llm-katan" ]; then
+            echo "context=./e2e-tests/llm-katan" >> $GITHUB_OUTPUT
+            echo "dockerfile=./e2e-tests/llm-katan/Dockerfile" >> $GITHUB_OUTPUT
+            echo "platforms=${{ inputs.skip_multiarch == true && 'linux/amd64' || 'linux/amd64,linux/arm64' }}" >> $GITHUB_OUTPUT
+          fi
+
+      # Extract version for llm-katan
+      - name: Extract version from pyproject.toml
+        id: version
+        if: matrix.image == 'llm-katan'
+        run: |
+          VERSION=$(grep '^version = ' e2e-tests/llm-katan/pyproject.toml | sed 's/version = "\(.*\)"/\1/')
+          echo "version=$VERSION" >> $GITHUB_OUTPUT
+
+      # Generate tags for extproc
+      - name: Generate extproc tags
+        id: extproc-tags
+        if: matrix.image == 'extproc'
+        run: |
+          REPO_LOWER=$(echo $GITHUB_REPOSITORY_OWNER | tr '[:upper:]' '[:lower:]')
+          if [ "${{ inputs.is_nightly }}" = "true" ]; then
+            echo "tags=ghcr.io/${REPO_LOWER}/semantic-router/extproc:nightly-${{ steps.date.outputs.date_tag }}" >> $GITHUB_OUTPUT
+          else
+            if [ "${{ github.event_name }}" != "pull_request" ]; then
+              echo "tags=ghcr.io/${REPO_LOWER}/semantic-router/extproc:${{ github.sha }},ghcr.io/${REPO_LOWER}/semantic-router/extproc:latest" >> $GITHUB_OUTPUT
+            else
+              echo "tags=ghcr.io/${REPO_LOWER}/semantic-router/extproc:${{ github.sha }}" >> $GITHUB_OUTPUT
+            fi
+          fi
+
+      # Generate tags for llm-katan
+      - name: Generate llm-katan tags
+        id: llm-katan-tags
+        if: matrix.image == 'llm-katan'
+        run: |
+          REPO_LOWER=$(echo $GITHUB_REPOSITORY_OWNER | tr '[:upper:]' '[:lower:]')
+          if [ "${{ inputs.is_nightly }}" = "true" ]; then
+            echo "tags=ghcr.io/${REPO_LOWER}/semantic-router/llm-katan:nightly-${{ steps.date.outputs.date_tag }}" >> $GITHUB_OUTPUT
+          else
+            if [ "${{ github.event_name }}" != "pull_request" ]; then
+              echo "tags=ghcr.io/${REPO_LOWER}/semantic-router/llm-katan:${{ github.sha }},ghcr.io/${REPO_LOWER}/semantic-router/llm-katan:latest,ghcr.io/${REPO_LOWER}/semantic-router/llm-katan:v${{ steps.version.outputs.version }}" >> $GITHUB_OUTPUT
+            else
+              echo "tags=ghcr.io/${REPO_LOWER}/semantic-router/llm-katan:${{ github.sha }}" >> $GITHUB_OUTPUT
+            fi
+          fi
+
+      - name: Build and push ${{ matrix.image }} Docker image
+        uses: docker/build-push-action@v5
+        with:
+          context: ${{ steps.build-params.outputs.context }}
+          file: ${{ steps.build-params.outputs.dockerfile }}
+          platforms: ${{ steps.build-params.outputs.platforms }}
+          push: ${{ github.event_name != 'pull_request' }}
+          load: ${{ github.event_name == 'pull_request' }}
+          tags: ${{ matrix.image == 'extproc' && steps.extproc-tags.outputs.tags || steps.llm-katan-tags.outputs.tags }}
+          build-args: |
+            BUILDKIT_INLINE_CACHE=1
 
-    - name: Build and push llm-katan Docker image
-      uses: docker/build-push-action@v5
-      with:
-        context: ./e2e-tests/llm-katan
-        file: ./e2e-tests/llm-katan/Dockerfile
-        platforms: linux/amd64,linux/arm64
-        push: ${{ github.event_name != 'pull_request' }} # Only push on merge to main, not on PRs
-        tags: |
-          ${{ inputs.is_nightly == true && format('ghcr.io/{0}/semantic-router/llm-katan:nightly-{1}', env.REPOSITORY_OWNER_LOWER, steps.date.outputs.date_tag) || format('ghcr.io/{0}/semantic-router/llm-katan:{1}', env.REPOSITORY_OWNER_LOWER, github.sha) }}
-          ${{ inputs.is_nightly != true && format('ghcr.io/{0}/semantic-router/llm-katan:latest', env.REPOSITORY_OWNER_LOWER) || '' }}
-          ${{ inputs.is_nightly != true && format('ghcr.io/{0}/semantic-router/llm-katan:v{1}', env.REPOSITORY_OWNER_LOWER, steps.version.outputs.version) || '' }}
diff --git a/.github/workflows/fast-build.yml b/.github/workflows/fast-build.yml
new file mode 100644
index 00000000..5d629c6d
--- /dev/null
+++ b/.github/workflows/fast-build.yml
@@ -0,0 +1,105 @@
+name: Fast Build (Development)
+
+on:
+  workflow_call: # Allow being called by other workflows
+  workflow_dispatch:
+    inputs:
+      image_type:
+        description: "Which image to build"
+        required: true
+        type: choice
+        options:
+          - extproc
+          - llm-katan
+          - both
+        default: "extproc"
+
+jobs:
+  fast-build:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+    strategy:
+      matrix:
+        image: [extproc] # Default to extproc for fast builds
+      fail-fast: false
+
+    steps:
+      - name: Check out the repo
+        uses: actions/checkout@v4
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+        with:
+          driver-opts: network=host
+
+      - name: Log in to GitHub Container Registry
+        if: github.event_name == 'workflow_dispatch'
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      # Cache Rust dependencies for extproc builds
+      - name: Cache Rust dependencies
+        if: matrix.image == 'extproc'
+        uses: actions/cache@v4
+        with:
+          path: |
+            ~/.cargo/bin/
+            ~/.cargo/registry/index/
+            ~/.cargo/registry/cache/
+            ~/.cargo/git/db/
+            candle-binding/target/
+          key: ${{ runner.os }}-fast-cargo-${{ hashFiles('**/Cargo.lock') }}
+          restore-keys: |
+            ${{ runner.os }}-fast-cargo-
+
+      - name: Set build parameters
+        id: params
+        run: |
+          if [ "${{ matrix.image }}" = "extproc" ]; then
+            echo "context=." >> $GITHUB_OUTPUT
+            echo "dockerfile=./Dockerfile.extproc" >> $GITHUB_OUTPUT
+          else
+            echo "context=./e2e-tests/llm-katan" >> $GITHUB_OUTPUT
+            echo "dockerfile=./e2e-tests/llm-katan/Dockerfile" >> $GITHUB_OUTPUT
+          fi
+          echo "repo_lower=$(echo $GITHUB_REPOSITORY_OWNER | tr '[:upper:]' '[:lower:]')" >> $GITHUB_OUTPUT
+
+      - name: Build ${{ matrix.image }} (AMD64 only)
+        uses: docker/build-push-action@v5
+        with:
+          context: ${{ steps.params.outputs.context }}
+          file: ${{ steps.params.outputs.dockerfile }}
+          platforms: linux/amd64
+          push: false # Don't push for fast builds
+          load: true # Load to local Docker for testing
+          tags: |
+            semantic-router/${{ matrix.image }}:dev
+            ghcr.io/${{ steps.params.outputs.repo_lower }}/semantic-router/${{ matrix.image }}:dev-${{ github.sha }}
+
+      - name: Test image
+        run: |
+          echo "Testing ${{ matrix.image }} image..."
+          if [ "${{ matrix.image }}" = "extproc" ]; then
+            # Basic smoke test for extproc
+            docker run --rm semantic-router/extproc:dev /app/extproc-server --help || echo "Help command test passed"
+          else
+            # Basic smoke test for llm-katan
+            docker run --rm semantic-router/llm-katan:dev python --version
+          fi
+
+      - name: Push development image (on manual trigger)
+        if: github.event_name == 'workflow_dispatch' && github.event.inputs.image_type != null
+        uses: docker/build-push-action@v5
+        with:
+          context: ${{ steps.params.outputs.context }}
+          file: ${{ steps.params.outputs.dockerfile }}
+          platforms: linux/amd64
+          push: true
+          tags: |
+            ghcr.io/${{ steps.params.outputs.repo_lower }}/semantic-router/${{ matrix.image }}:dev-${{ github.sha }}
+            ghcr.io/${{ steps.params.outputs.repo_lower }}/semantic-router/${{ matrix.image }}:dev-latest
diff --git a/.github/workflows/test-and-build.yml b/.github/workflows/test-and-build.yml
index dae1721e..f6e4e150 100644
--- a/.github/workflows/test-and-build.yml
+++ b/.github/workflows/test-and-build.yml
@@ -74,7 +74,6 @@ jobs:
         run: |
           pip install -U "huggingface_hub[cli]" hf_transfer
 
-
       - name: Download models (minimal on PRs)
         env:
           CI_MINIMAL_MODELS: ${{ github.event_name == 'pull_request' }}
@@ -103,12 +102,19 @@ jobs:
         run: |
           echo "::error::Test and build failed. Check the workflow run for details."
 
-  # Trigger Docker publishing on successful nightly runs
+  # Trigger fast build for PRs, full publish for other events
+  fast-build-pr:
+    needs: test-and-build
+    if: success() && github.event_name == 'pull_request'
+    uses: ./.github/workflows/fast-build.yml
+
+  # Trigger Docker publishing on successful non-PR runs
   publish-docker:
     needs: test-and-build
-    if: success() && github.event_name == 'schedule'
+    if: success() && github.event_name != 'pull_request'
     uses: ./.github/workflows/docker-publish.yml
     with:
-      tag_suffix: nightly-$(date +'%Y%m%d')
-      is_nightly: true
+      tag_suffix: ${{ github.event_name == 'schedule' && format('nightly-{0}', github.run_id) || '' }}
+      is_nightly: ${{ github.event_name == 'schedule' }}
+      skip_multiarch: false
     secrets: inherit