feat(component): add ChatLLM binding for chat conversations

- Add ChatLLM binding to support separate LLM model for chat
- Update CallLLMNode to use ChatLLM instead of CheapLLM
- Update test fixtures to bind ChatLLM in all configurations
- Update unit tests to include ChatLLM binding

This enables 3 distinct LLM models:
- SmartLLM: Query generation from scratch
- CheapLLM: Query generation from cached samples
- ChatLLM: Chat conversations

Author: sf-sahil-jassal

package-lock.json

@@ -44,20 +44,23 @@ export class TestApp extends BootMixin(
       this.bind(AiIntegrationBindings.CheapLLM).toProvider(Ollama);
       this.bind(AiIntegrationBindings.SmartLLM).toProvider(Ollama);
       this.bind(AiIntegrationBindings.FileLLM).toProvider(Ollama);
+      this.bind(AiIntegrationBindings.ChatLLM).toProvider(Ollama);
       this.bind(AiIntegrationBindings.EmbeddingModel).toProvider(
         OllamaEmbedding,
       );
     } else if (process.env.CEREBRAS === '1') {
       this.bind(AiIntegrationBindings.CheapLLM).toProvider(Cerebras);
       this.bind(AiIntegrationBindings.SmartLLM).toProvider(Cerebras);
       this.bind(AiIntegrationBindings.FileLLM).toProvider(Cerebras);
+      this.bind(AiIntegrationBindings.ChatLLM).toProvider(Cerebras);
       this.bind(AiIntegrationBindings.EmbeddingModel).toProvider(
         OllamaEmbedding,
       );
     } else if (options.llmStub) {
       this.bind(AiIntegrationBindings.CheapLLM).to(options.llmStub);
       this.bind(AiIntegrationBindings.SmartLLM).to(options.llmStub);
       this.bind(AiIntegrationBindings.FileLLM).to(options.llmStub);
+      this.bind(AiIntegrationBindings.ChatLLM).to(options.llmStub);
       this.bind(AiIntegrationBindings.EmbeddingModel).to(options.llmStub);
     }
     this.bind('datasources.readerdb').to(

src/__tests__/fixtures/test-app.ts

@@ -37,6 +37,7 @@ describe('CallLLMNode Unit', function () {
     context.bind(AuthenticationBindings.CURRENT_USER).to(stubUser());
     context.bind(AiIntegrationBindings.SmartLLM).to(llmProvider);
     context.bind(AiIntegrationBindings.CheapLLM).to(llmProvider);
+    context.bind(AiIntegrationBindings.ChatLLM).to(llmProvider);
     context.bind('datasources.readerdb').to(
       new juggler.DataSource({
         connector: 'sqlite3',

src/__tests__/unit/nodes/call-llm.node.unit.ts

@@ -18,9 +18,10 @@ import {Concept, GraphEdge, GraphNode, KnowledgeGraph} from './types';
 const debug = require('debug')('ai-integration:knowledge-graph');
 
 @injectable({scope: BindingScope.SINGLETON})
-export class DbKnowledgeGraphService
-  implements KnowledgeGraph<string, DatabaseSchema>
-{
+export class DbKnowledgeGraphService implements KnowledgeGraph<
+  string,
+  DatabaseSchema
+> {
   edges: Map<string, GraphEdge[]>;
   nodes: Map<string, GraphNode>;
   private vectorWeight: number;

src/components/db-query/services/knowledge-graph/db-knowledge-graph.service.ts

@@ -8,9 +8,7 @@ import {VisualizationGraphState} from '../state';
 @graphNode(VisualizationGraphNodes.CallQueryGeneration, {
   [POST_DATASET_TAG]: true,
 })
-export class CallQueryGenerationNode
-  implements IGraphNode<VisualizationGraphState>
-{
+export class CallQueryGenerationNode implements IGraphNode<VisualizationGraphState> {
   constructor(
     @service(DbQueryGraph)
     private readonly queryPipeline: DbQueryGraph,

src/components/visualization/nodes/call-query-generation.node.ts

@@ -8,9 +8,7 @@ import {POST_DATASET_TAG} from '../../db-query';
 @graphNode(VisualizationGraphNodes.RenderVisualization, {
   [POST_DATASET_TAG]: true,
 })
-export class RenderVisualizationNode
-  implements IGraphNode<VisualizationGraphState>
-{
+export class RenderVisualizationNode implements IGraphNode<VisualizationGraphState> {
   constructor() {}
 
   async execute(

src/components/visualization/nodes/render-visualization.node.ts

@@ -15,9 +15,7 @@ import {POST_DATASET_TAG} from '../../db-query';
 @graphNode(VisualizationGraphNodes.SelectVisualisation, {
   [POST_DATASET_TAG]: true,
 })
-export class SelectVisualizationNode
-  implements IGraphNode<VisualizationGraphState>
-{
+export class SelectVisualizationNode implements IGraphNode<VisualizationGraphState> {
   prompt = PromptTemplate.fromTemplate(`
 <instructions>
 You are expert Data Analysis Agent whose job is to suggest visualisations that would be best suited to display the results for a particular user prompt and the data extracted based on that prompt.

src/components/visualization/nodes/select-visualization.node.ts

@@ -17,7 +17,7 @@ const debug = require('debug')('ai-integration:chat:call-llm.node');
 @graphNode(ChatNodes.CallLLM)
 export class CallLLMNode implements IGraphNode<ChatState> {
   constructor(
-    @inject(AiIntegrationBindings.CheapLLM)
+    @inject(AiIntegrationBindings.ChatLLM)
     private readonly llm: LLMProvider,
     @inject(AiIntegrationBindings.Tools)
     private readonly tools: ToolStore,

src/graphs/chat/nodes/call-llm.node.ts

@@ -24,6 +24,9 @@ export namespace AiIntegrationBindings {
   export const FileLLM = BindingKey.create<LLMProvider>(
     'services.ai-reporting.fileLLMProvider',
   );
+  export const ChatLLM = BindingKey.create<LLMProvider>(
+    'services.ai-reporting.chatLLMProvider',
+  );
   export const EmbeddingModel = BindingKey.create<EmbeddingProvider>(
     'services.ai-reporting.embeddingModel',
   );