Add chunking and multi-query retrieval

backend/src/ingestion_graph/graph.ts

@@ -4,6 +4,7 @@
 
 import { RunnableConfig } from '@langchain/core/runnables';
 import { StateGraph, END, START } from '@langchain/langgraph';
+import { RecursiveCharacterTextSplitter } from '@langchain/textsplitters';
 import fs from 'fs/promises';
 
 import { IndexStateAnnotation } from './state.js';
@@ -14,6 +15,11 @@ import {
 } from './configuration.js';
 import { reduceDocs } from '../shared/state.js';
 
+const textSplitter = new RecursiveCharacterTextSplitter({
+  chunkSize: 1000,
+  chunkOverlap: 200,
+});
+
 async function ingestDocs(
   state: typeof IndexStateAnnotation.State,
   config?: RunnableConfig,
@@ -37,8 +43,11 @@ async function ingestDocs(
     docs = reduceDocs([], docs);
   }
 
+  // Split documents into smaller chunks for better retrieval
+  const splitDocs = await textSplitter.splitDocuments(docs);
+
   const retriever = await makeRetriever(config);
-  await retriever.addDocuments(docs);
+  await retriever.addDocuments(splitDocs);
 
   return { docs: 'delete' };
 }

backend/src/retrieval_graph/graph.ts

@@ -4,7 +4,11 @@ import { makeRetriever } from '../shared/retrieval.js';
 import { formatDocs } from './utils.js';
 import { HumanMessage } from '@langchain/core/messages';
 import { z } from 'zod';
-import { RESPONSE_SYSTEM_PROMPT, ROUTER_SYSTEM_PROMPT } from './prompts.js';
+import {
+  RESPONSE_SYSTEM_PROMPT,
+  ROUTER_SYSTEM_PROMPT,
+  MULTI_QUERY_PROMPT,
+} from './prompts.js';
 import { RunnableConfig } from '@langchain/core/runnables';
 import {
   AgentConfigurationAnnotation,
@@ -74,10 +78,40 @@ async function retrieveDocuments(
   state: typeof AgentStateAnnotation.State,
   config: RunnableConfig,
 ): Promise<typeof AgentStateAnnotation.Update> {
+  const configuration = ensureAgentConfiguration(config);
+  const model = await loadChatModel(configuration.queryModel);
   const retriever = await makeRetriever(config);
-  const response = await retriever.invoke(state.query);
 
-  return { documents: response };
+  // Generate multiple query variants for better retrieval coverage
+  const formattedPrompt = await MULTI_QUERY_PROMPT.invoke({
+    query: state.query,
+  });
+  const queryResponse = await model.invoke(formattedPrompt.toString());
+  const queries = [
+    state.query,
+    ...String(queryResponse.content)
+      .split('\n')
+      .map((q) => q.trim())
+      .filter((q) => q.length > 0),
+  ];
+
+  // Retrieve documents for all queries in parallel
+  const allResults = await Promise.all(
+    queries.map((q) => retriever.invoke(q)),
+  );
+
+  // Deduplicate by page content, preserving order (original query first)
+  const seen = new Set<string>();
+  const uniqueDocs = allResults.flat().filter((doc) => {
+    const key = doc.pageContent;
+    if (seen.has(key)) return false;
+    seen.add(key);
+    return true;
+  });
+
+  // Cap results — original query docs are first, so most relevant are kept
+  const maxDocs = configuration.k * 2;
+  return { documents: uniqueDocs.slice(0, maxDocs) };
 }
 
 async function generateResponse(

backend/src/retrieval_graph/prompts.ts

@@ -23,4 +23,16 @@ const RESPONSE_SYSTEM_PROMPT = ChatPromptTemplate.fromMessages([
   ],
 ]);
 
-export { ROUTER_SYSTEM_PROMPT, RESPONSE_SYSTEM_PROMPT };
+const MULTI_QUERY_PROMPT = ChatPromptTemplate.fromMessages([
+  [
+    'system',
+    `You are an AI assistant that generates multiple search queries to improve document retrieval.
+Given a user question, generate 3 different versions of the question that capture different aspects or phrasings.
+Each query should approach the question from a different angle to maximize the chance of finding relevant documents.
+
+Return ONLY the 3 queries, one per line, with no numbering or prefixes.`,
+  ],
+  ['human', '{query}'],
+]);
+
+export { ROUTER_SYSTEM_PROMPT, RESPONSE_SYSTEM_PROMPT, MULTI_QUERY_PROMPT };

frontend/components/chat-message.tsx

@@ -21,6 +21,7 @@ interface ChatMessageProps {
 export function ChatMessage({ message }: ChatMessageProps) {
   const isUser = message.role === 'user';
   const [copied, setCopied] = useState(false);
+  const [expandedSource, setExpandedSource] = useState<number | null>(null);
   const isLoading = message.role === 'assistant' && message.content === '';
 
   const handleCopy = async () => {
@@ -78,17 +79,30 @@ export function ChatMessage({ message }: ChatMessageProps) {
                       {message.sources?.map((source, index) => (
                         <Card
                           key={index}
-                          className="bg-background/50 transition-all duration-200 hover:bg-background hover:shadow-md hover:scale-[1.02] cursor-pointer"
+                          className="bg-background/50 transition-all duration-200 hover:bg-background hover:shadow-md cursor-pointer"
+                          onClick={() =>
+                            setExpandedSource(
+                              expandedSource === index ? null : index,
+                            )
+                          }
                         >
                           <CardContent className="p-3">
                             <p className="text-sm font-medium truncate">
-                              {source.metadata?.source ||
-                                source.metadata?.filename ||
-                                'N/A'}
+                              {source.metadata?.filename ||
+                                (source.metadata?.source
+                                  ? source.metadata.source
+                                      .split(/[/\\]/)
+                                      .pop()
+                                  : 'N/A')}
                             </p>
                             <p className="text-sm text-muted-foreground">
                               Page {source.metadata?.loc?.pageNumber || 'N/A'}
                             </p>
+                            {expandedSource === index && (
+                              <p className="text-xs text-muted-foreground mt-2 whitespace-pre-wrap border-t pt-2">
+                                {source.pageContent || 'No content available'}
+                              </p>
+                            )}
                           </CardContent>
                         </Card>
                       ))}