feat: implement decision-based routing with plugin architecture

Signed-off-by: bitliu <[email protected]>

Author: Xunzhuo

deploy/kubernetes/crds/examples/intelligentpool-example.yaml

@@ -36,4 +36,3 @@ spec:
       pricing:
         inputTokenPrice: 0.0000005
         outputTokenPrice: 0.000001
-

deploy/kubernetes/crds/examples/intelligentroute-example.yaml

@@ -4,50 +4,54 @@ metadata:
   name: default-route
   namespace: default
 spec:
-  strategy: "priority"  # or "confidence"
-  
+  decisionStrategy: "priority"  # or "confidence"
+
   # Keyword Rules
   keywordRules:
     - name: "urgent_keyword"
-      category: "urgent request"
       operator: "OR"
       keywords: ["urgent", "emergency", "asap", "critical"]
       caseSensitive: false
-    
+
     - name: "technical_keyword"
-      category: "technical support"
       operator: "AND"
       keywords: ["technical", "support"]
       caseSensitive: false
-  
+
   # Embedding Rules
   embeddingRules:
     - name: "tech_support_embedding"
-      category: "technical_support"
       threshold: 0.72
-      keywords: ["technical support", "troubleshooting", "debugging"]
+      candidates:
+        - "I need help with technical issues and troubleshooting"
+        - "Can you help me debug this problem?"
+        - "Technical support for system errors"
       aggregationMethod: "max"
-      model: "auto"
-      dimension: 768
-    
+
     - name: "sales_embedding"
-      category: "sales_inquiry"
       threshold: 0.75
-      keywords: ["pricing", "purchase", "buy", "sales"]
+      candidates:
+        - "I want to know about pricing and purchase options"
+        - "How can I buy this product?"
+        - "Tell me about sales and discounts"
       aggregationMethod: "mean"
-      model: "auto"
-      dimension: 768
-  
-  # Domain Rules
+
+  # Domain Rules (MMLU domain categories)
   domainRules:
-    - name: "math_domain"
-      domains: ["math", "physics"]
-    
-    - name: "science_domain"
-      domains: ["biology", "chemistry", "physics"]
-    
-    - name: "business_domain"
-      domains: ["business", "economics"]
+    - "business"
+    - "law"
+    - "psychology"
+    - "biology"
+    - "chemistry"
+    - "history"
+    - "other"
+    - "health"
+    - "economics"
+    - "math"
+    - "physics"
+    - "computer_science"
+    - "philosophy"
+    - "engineering"
 
   # Decisions
   decisions:
@@ -100,7 +104,7 @@ spec:
         operator: "OR"
         conditions:
           - type: "domain"
-            name: "math_domain"
+            name: "math"
       modelRefs:
         - model: "base-model"
           loraName: "math-expert"
@@ -149,4 +153,3 @@ spec:
           configuration:
             enabled: true
             threshold: 0.5
-

deploy/kubernetes/crds/intelligentpool-crd.yaml

@@ -118,4 +118,3 @@ spec:
         - name: Age
           type: date
           jsonPath: .metadata.creationTimestamp
-

deploy/kubernetes/crds/intelligentroute-crd.yaml

@@ -23,10 +23,10 @@ spec:
             spec:
               type: object
               required:
-                - strategy
+                - decisionStrategy
                 - decisions
               properties:
-                strategy:
+                decisionStrategy:
                   type: string
                   enum:
                     - priority
@@ -38,16 +38,12 @@ spec:
                     type: object
                     required:
                       - name
-                      - category
                       - operator
                       - keywords
                     properties:
                       name:
                         type: string
-                        description: Unique name for this rule
-                      category:
-                        type: string
-                        description: Category name
+                        description: Unique name for this rule (also used as category name)
                       operator:
                         type: string
                         enum:
@@ -67,67 +63,46 @@ spec:
                     type: object
                     required:
                       - name
-                      - category
                       - threshold
-                      - keywords
+                      - candidates
                     properties:
                       name:
                         type: string
-                        description: Unique name for this rule
-                      category:
-                        type: string
-                        description: Category name
+                        description: Unique name for this rule (also used as category name)
                       threshold:
                         type: number
                         format: float
                         description: Similarity threshold
-                      keywords:
+                      candidates:
                         type: array
                         items:
                           type: string
+                        description: Candidate phrases for semantic matching
                       aggregationMethod:
                         type: string
                         enum:
                           - max
                           - mean
                         default: max
-                      model:
-                        type: string
-                        default: auto
-                      dimension:
-                        type: integer
-                        format: int32
-                        default: 768
                 domainRules:
                   type: array
                   items:
-                    type: object
-                    required:
-                      - name
-                      - domains
-                    properties:
-                      name:
-                        type: string
-                        description: Unique name for this rule
-                      domains:
-                        type: array
-                        items:
-                          type: string
-                          enum:
-                            - business
-                            - law
-                            - psychology
-                            - biology
-                            - chemistry
-                            - history
-                            - other
-                            - health
-                            - economics
-                            - math
-                            - physics
-                            - computer science
-                            - philosophy
-                            - engineering
+                    type: string
+                    enum:
+                      - business
+                      - law
+                      - psychology
+                      - biology
+                      - chemistry
+                      - history
+                      - other
+                      - health
+                      - economics
+                      - math
+                      - physics
+                      - computer_science
+                      - philosophy
+                      - engineering
                 decisions:
                   type: array
                   items:
@@ -276,4 +251,3 @@ spec:
         - name: Age
           type: date
           jsonPath: .metadata.creationTimestamp
-