fixed race condition between the RAG pipeline and AI response parsing and replaced sonnet with Haiku

Server/app/worker-service/services/rag.service.ts

@@ -308,6 +308,24 @@ export async function vectorDBExists(DBName: string): Promise<boolean> {
   return indexes?.some((idx) => idx.name === DBName) ?? false;
 }
 
+const sleep = (ms: number) => new Promise((resolve) => setTimeout(resolve, ms));
+
+export async function waitForVectorIndex(
+  indexName: string,
+  options: { timeoutMs?: number; intervalMs?: number } = {},
+): Promise<boolean> {
+  const { timeoutMs = 10 * 60 * 1000, intervalMs = 5000 } = options;
+  const deadline = Date.now() + timeoutMs;
+
+  while (Date.now() < deadline) {
+    if (await vectorDBExists(indexName)) return true;
+    console.log(`Waiting for vector index "${indexName}"...`);
+    await sleep(intervalMs);
+  }
+
+  return false;
+}
+
 export async function runRAGPipeline(data: object) {
   const { installationId, owner, repo } = data as RAGPipelineInput;
 

Server/app/worker-service/services/review.service.ts

@@ -15,9 +15,8 @@ import {
 import { db } from "../../../package/db/prisma.js";
 import {
   vectorDBExists,
-  createEmbeddings,
-  saveToVectorDB,
-  searchRelevantEmbeddings,
+  waitForVectorIndex,
+  retrieveContextForDiff,
   toIndexName,
 } from "./rag.service.js";
 import { findGIF } from "./gif.service.js";
@@ -137,17 +136,6 @@ async function resolveUserId(
 
 export async function runPRReview(data: PRReviewJobData) {
   const { installationId, owner, repo, prNumber, prTitle, prAuthor } = data;
-  const dbName = "";
-
-  const createVectorDB = (await vectorDBExists(dbName)) as boolean;
-
-  if (createVectorDB) {
-    const values = {};
-
-    const embeddings = (await createEmbeddings(values)) as any;
-    const finalSaved = await saveToVectorDB(embeddings);
-    console.log(finalSaved);
-  }
 
   const octokit = await getOctokit(installationId);
   const userId = await resolveUserId(owner, repo, data.userId);
@@ -180,17 +168,34 @@ export async function runPRReview(data: PRReviewJobData) {
   const rules = await getReviewRules(octokit, owner, repo, prNumber);
   const reviewType = (await getRevieType(difference, apiKey, tokenUsage)) as any;
   const indexName = toIndexName(owner, repo);
-  const DBExsists = await vectorDBExists(indexName);
-  let relevantCode: any;
+  let relevantCode: Awaited<ReturnType<typeof retrieveContextForDiff>> | undefined;
   const processedDiff = (await getCodeDiff(difference)) as any;
-  if (reviewType === "feature" && DBExsists) {
-    const searchQuery = await generateRelevantSearchQuery(processedDiff, apiKey, tokenUsage);
-    const query = {
-      text: searchQuery,
-      indexName: indexName,
-      topK: 5,
-    };
-    relevantCode = await searchRelevantEmbeddings(query);
+
+  if (reviewType === "feature") {
+    const indexReady =
+      (await vectorDBExists(indexName)) ||
+      (await waitForVectorIndex(indexName));
+
+    if (indexReady) {
+      const searchQuery = await generateRelevantSearchQuery(
+        processedDiff,
+        apiKey,
+        tokenUsage,
+      );
+      relevantCode = await retrieveContextForDiff({
+        diff: difference,
+        semanticQuery: searchQuery,
+        indexName,
+        topK: 8,
+      });
+      console.log(
+        `RAG context for ${owner}/${repo}#${prNumber}: ${relevantCode.length} AST chunks`,
+      );
+    } else {
+      console.log(
+        `Vector index "${indexName}" not ready — reviewing without RAG context`,
+      );
+    }
   }
 
   const prompt = reviewPrompt(difference, rules, relevantCode);

Server/app/worker-service/workers/review.worker.ts

@@ -5,8 +5,12 @@ export async function workersOn(){
     const client = await createWorkerRedisClient()
     console.log("webhook for PR is on")
 
-    while(true){
-        const gettingData = await client.brPop("reviewQueue", 0)
-        processPRReviewJob(JSON.parse(gettingData!.element))
+    while (true) {
+        const gettingData = await client.brPop("reviewQueue", 0);
+        try {
+            await processPRReviewJob(JSON.parse(gettingData!.element));
+        } catch (err) {
+            console.error("PR review job error:", err);
+        }
     }
 }

Server/package/ai/deployment.ts

@@ -1,12 +1,13 @@
 import Anthropic from "@anthropic-ai/sdk";
+import { CLAUDE_MODEL } from "./models.js";
 
 const anthropic = new Anthropic({
   apiKey: process.env.ANTHROPIC_API_KEY!,
 });
 
 export async function analyzeDeploymentLogs(logs: string, provider: string) {
   const res = await anthropic.messages.create({
-    model: "claude-sonnet-4-6",
+    model: CLAUDE_MODEL,
     max_tokens: 1500,
     system: `You are a senior DevOps engineer analyzing CI/CD deployment failure logs.
 
@@ -44,7 +45,7 @@ Do NOT include markdown or text outside the JSON.`,
 
 export async function generateDeploymentFix(fileContent: string, cause: string, fix: string, fileName: string) {
   const res = await anthropic.messages.create({
-    model: "claude-sonnet-4-6",
+    model: CLAUDE_MODEL,
     max_tokens: 4000,
     system: `You are a senior software engineer. You will be given a file and a description of a bug causing a deployment failure. Apply the fix and return the complete corrected file content.
 

Server/package/ai/gif.ts

@@ -1,5 +1,6 @@
 import Anthropic from "@anthropic-ai/sdk";
 import { addUsage, createTokenAccumulator, type TokenAccumulator } from "./review.js";
+import { CLAUDE_MODEL } from "./models.js";
 
 export async function getGifName(
   summary: string | null | undefined,
@@ -13,7 +14,7 @@ export async function getGifName(
 
   const anthropic = new Anthropic({ apiKey });
   const res = await anthropic.messages.create({
-    model: "claude-sonnet-4-5",
+    model: CLAUDE_MODEL,
     max_tokens: 40,
     temperature: 0,
     system:

Server/package/ai/models.ts

@@ -0,0 +1,2 @@
+export const CLAUDE_MODEL =
+  process.env.CLAUDE_MODEL?.trim() || "claude-haiku-4-5";

Server/package/ai/review.ts

@@ -1,4 +1,5 @@
 import Anthropic from "@anthropic-ai/sdk";
+import { CLAUDE_MODEL } from "./models.js";
 
 export type TokenAccumulator = {
   inputTokens: number;
@@ -21,15 +22,16 @@ function createClient(apiKey: string) {
 
 export async function getRevieType(diff: string, apiKey: string, usage?: TokenAccumulator) {
   const anthropic = createClient(apiKey);
-  const prompt = `You are supposed to see the Code Diff ${diff} and identify the PR Type if it is a Bug Fix / New Feature / Code Update , etc and only return the name of the type in response . "Return ONLY: feature | bugfix | refactor"`;
+  const diffPreview = diff.slice(0, 3000);
+  const prompt = `Identify the PR type from this code diff snippet. Return ONLY one word: feature | bugfix | refactor\n\n${diffPreview}`;
   const res = await anthropic.messages.create({
-    model: "claude-sonnet-4-5",
-    max_tokens: 1500,
+    model: CLAUDE_MODEL,
+    max_tokens: 16,
     temperature: 0,
     messages: [
       {
         role: "user",
-        content: prompt ? prompt : "",
+        content: prompt,
       },
     ],
   });
@@ -53,8 +55,8 @@ export async function generateRelevantSearchQuery(
 ) {
   const anthropic = createClient(apiKey);
   const res = await anthropic.messages.create({
-    model: "claude-sonnet-4-5",
-    max_tokens: 1500,
+    model: CLAUDE_MODEL,
+    max_tokens: 100,
     temperature: 0,
     system:
       "Convert the given git diff into a short semantic search query  to find relevant code in a repository Return ONLY a short phrase.",
@@ -97,8 +99,8 @@ export async function getAIReview(
 async function _callAI(prompt: string | null, apiKey: string, usage?: TokenAccumulator) {
   const anthropic = createClient(apiKey);
   const res = await anthropic.messages.create({
-    model: "claude-sonnet-4-5",
-    max_tokens: 1500,
+    model: CLAUDE_MODEL,
+    max_tokens: 4096,
     temperature: 0,
     system: `You are a strict senior software engineer performing code reviews.
 
@@ -167,6 +169,10 @@ Do NOT include markdown, explanations, or text outside JSON.`,
   });
   addUsage(usage ?? createTokenAccumulator(), res.usage);
 
+  if (res.stop_reason === "max_tokens") {
+    console.warn("AI review response was truncated (max_tokens reached)");
+  }
+
   const block = res.content.find((b) => b.type === "text");
   if (!block || block.type !== "text")
     throw new Error("No text block in AI response");
@@ -199,6 +205,8 @@ function formatRelevantCode(matches: RelevantCodeMatch[]): string {
     .join("\n\n---\n\n")
 }
 
+const MAX_DIFF_CHARS = 80_000;
+
 export function reviewPrompt(
   diff: string,
   rules: any,
@@ -209,6 +217,11 @@ export function reviewPrompt(
       ? `\nEXISTING CODEBASE CONTEXT (retrieved for this feature):\nUse this to check for consistency, duplication, or conflicts with existing patterns.\n\n${formatRelevantCode(relevantCode)}\n`
       : "";
 
+  const truncatedDiff =
+    diff.length > MAX_DIFF_CHARS
+      ? `${diff.slice(0, MAX_DIFF_CHARS)}\n\n[... diff truncated at ${MAX_DIFF_CHARS} chars ...]`
+      : diff;
+
   return `
 REPOSITORY RULES (HIGHEST PRIORITY):
 ${JSON.stringify(rules, null, 2)}
@@ -222,19 +235,37 @@ Check for:
 - Maintainability (naming, duplication)
 ${relevantSection}
 GIT DIFF:
-${diff}
+${truncatedDiff}
 
 Analyze ONLY the changed code.
 
 Return strictly valid JSON.
 `;
 }
 
+function extractJsonObject(text: string): string | null {
+  const fenced = text.match(/```(?:json)?\s*([\s\S]*?)```/i);
+  if (fenced?.[1]) return fenced[1].trim();
+
+  const start = text.indexOf("{");
+  const end = text.lastIndexOf("}");
+  if (start !== -1 && end > start) return text.slice(start, end + 1);
+
+  return null;
+}
+
 export function parseAIResponse(text: string) {
-  try {
-    return JSON.parse(text);
-  } catch (err) {
-    console.log("❌ Failed to parse AI response");
-    return null;
+  const candidates = [text.trim(), extractJsonObject(text)].filter(Boolean) as string[];
+
+  for (const candidate of candidates) {
+    try {
+      return JSON.parse(candidate);
+    } catch {
+      // try next candidate
+    }
   }
+
+  console.log("❌ Failed to parse AI response");
+  console.log("Response preview:", text.slice(0, 500));
+  return null;
 }