@retrivora-ai/rag-engine 0.2.3 → 0.2.4

package/dist/{chunk-K53N7SEE.mjs → chunk-7K4KXB6G.mjs}

@@ -1640,7 +1640,7 @@ var Pipeline = class {
     const scoreThreshold = (_d = (_c = this.config.rag) == null ? void 0 : _c.scoreThreshold) != null ? _d : 0;
     try {
       const queryVector = await this.embeddingProvider.embed(question, { taskType: "query" });
-      const rawMatches = await this.vectorDB.query(queryVector, topK, ns);
+      const rawMatches = await this.vectorDB.query(queryVector, topK, ns, { __queryText: question });
       const sources = rawMatches.filter((m) => m.score >= scoreThreshold);
       const context = sources.length ? sources.map((m, i) => `[Source ${i + 1}]
 ${m.content}`).join("\n\n---\n\n") : "No relevant context found.";

package/dist/handlers/index.js

@@ -2750,7 +2750,7 @@ var Pipeline = class {
     const scoreThreshold = (_d = (_c = this.config.rag) == null ? void 0 : _c.scoreThreshold) != null ? _d : 0;
     try {
       const queryVector = await this.embeddingProvider.embed(question, { taskType: "query" });
-      const rawMatches = await this.vectorDB.query(queryVector, topK, ns);
+      const rawMatches = await this.vectorDB.query(queryVector, topK, ns, { __queryText: question });
       const sources = rawMatches.filter((m) => m.score >= scoreThreshold);
       const context = sources.length ? sources.map((m, i) => `[Source ${i + 1}]
 ${m.content}`).join("\n\n---\n\n") : "No relevant context found.";

package/dist/handlers/index.mjs

@@ -3,7 +3,7 @@ import {
   createHealthHandler,
   createIngestHandler,
   createUploadHandler
-} from "../chunk-K53N7SEE.mjs";
+} from "../chunk-7K4KXB6G.mjs";
 import "../chunk-EDLTMSNY.mjs";
 import "../chunk-FWCSY2DS.mjs";
 export {

package/dist/server.js

@@ -2839,7 +2839,7 @@ var Pipeline = class {
     const scoreThreshold = (_d = (_c = this.config.rag) == null ? void 0 : _c.scoreThreshold) != null ? _d : 0;
     try {
       const queryVector = await this.embeddingProvider.embed(question, { taskType: "query" });
-      const rawMatches = await this.vectorDB.query(queryVector, topK, ns);
+      const rawMatches = await this.vectorDB.query(queryVector, topK, ns, { __queryText: question });
       const sources = rawMatches.filter((m) => m.score >= scoreThreshold);
       const context = sources.length ? sources.map((m, i) => `[Source ${i + 1}]
 ${m.content}`).join("\n\n---\n\n") : "No relevant context found.";
@@ -3770,29 +3770,46 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
     const vectorLiteral = `[${vector.join(",")}]`;
     const allResults = [];
     console.log(`[MultiTablePostgresProvider] --- Starting Multi-Table Search ---`);
+    const queryText = _filter == null ? void 0 : _filter.__queryText;
     for (const table of this.tables) {
       try {
-        const result = await this.pool.query(
-          `SELECT *, 
-           (1 - (embedding <=> $1::vector)) AS vector_score
-           FROM "${table}"
-           ORDER BY vector_score DESC
-           LIMIT 50`,
-          [vectorLiteral]
-        );
+        let sqlQuery = "";
+        let params = [];
+        if (queryText) {
+          sqlQuery = `
+            SELECT *, 
+             (1 - (embedding <=> $1::vector)) AS vector_score,
+             ts_rank(to_tsvector('english', t.*::text), plainto_tsquery('english', $2)) AS keyword_score,
+             ((1 - (embedding <=> $1::vector)) + (ts_rank(to_tsvector('english', t.*::text), plainto_tsquery('english', $2)) * 2.0)) AS hybrid_score
+            FROM "${table}" t
+            ORDER BY hybrid_score DESC
+            LIMIT 50
+          `;
+          params = [vectorLiteral, queryText];
+        } else {
+          sqlQuery = `
+            SELECT *, 
+             (1 - (embedding <=> $1::vector)) AS hybrid_score
+            FROM "${table}" t
+            ORDER BY hybrid_score DESC
+            LIMIT 50
+          `;
+          params = [vectorLiteral];
+        }
+        const result = await this.pool.query(sqlQuery, params);
         if (result.rowCount && result.rowCount > 0) {
-          console.log(`   \u2705 Table "${table}": Found ${result.rowCount} potential matches. Top score: ${result.rows[0].vector_score.toFixed(4)}`);
+          console.log(`   \u2705 Table "${table}": Found ${result.rowCount} potential matches. Top score: ${result.rows[0].hybrid_score.toFixed(4)}`);
         } else {
           console.log(`   \u26AA Table "${table}": No relevant matches found.`);
         }
         for (const row of result.rows) {
-          const _a = row, { vector_score, id } = _a, rest = __objRest(_a, ["vector_score", "id"]);
+          const _a = row, { hybrid_score, vector_score, keyword_score, id } = _a, rest = __objRest(_a, ["hybrid_score", "vector_score", "keyword_score", "id"]);
           delete rest.embedding;
           const content = `[TYPE: ${table.replace(/s$/, "").toUpperCase()}]
 ` + Object.entries(rest).filter(([k, v]) => v !== null && typeof v !== "object" && k !== "id").map(([k, v]) => `${k}: ${v}`).join("\n");
           allResults.push({
             id: `${table}-${id}`,
-            score: parseFloat(String(vector_score)),
+            score: parseFloat(String(hybrid_score)),
             content,
             metadata: __spreadProps(__spreadValues({}, rest), { source_table: table })
           });

package/dist/server.mjs

@@ -34,7 +34,7 @@ import {
   createIngestHandler,
   createUploadHandler,
   getRagConfig
-} from "./chunk-K53N7SEE.mjs";
+} from "./chunk-7K4KXB6G.mjs";
 import "./chunk-EDLTMSNY.mjs";
 import {
   PineconeProvider
@@ -413,29 +413,46 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
     const vectorLiteral = `[${vector.join(",")}]`;
     const allResults = [];
     console.log(`[MultiTablePostgresProvider] --- Starting Multi-Table Search ---`);
+    const queryText = _filter == null ? void 0 : _filter.__queryText;
     for (const table of this.tables) {
       try {
-        const result = await this.pool.query(
-          `SELECT *, 
-           (1 - (embedding <=> $1::vector)) AS vector_score
-           FROM "${table}"
-           ORDER BY vector_score DESC
-           LIMIT 50`,
-          [vectorLiteral]
-        );
+        let sqlQuery = "";
+        let params = [];
+        if (queryText) {
+          sqlQuery = `
+            SELECT *, 
+             (1 - (embedding <=> $1::vector)) AS vector_score,
+             ts_rank(to_tsvector('english', t.*::text), plainto_tsquery('english', $2)) AS keyword_score,
+             ((1 - (embedding <=> $1::vector)) + (ts_rank(to_tsvector('english', t.*::text), plainto_tsquery('english', $2)) * 2.0)) AS hybrid_score
+            FROM "${table}" t
+            ORDER BY hybrid_score DESC
+            LIMIT 50
+          `;
+          params = [vectorLiteral, queryText];
+        } else {
+          sqlQuery = `
+            SELECT *, 
+             (1 - (embedding <=> $1::vector)) AS hybrid_score
+            FROM "${table}" t
+            ORDER BY hybrid_score DESC
+            LIMIT 50
+          `;
+          params = [vectorLiteral];
+        }
+        const result = await this.pool.query(sqlQuery, params);
         if (result.rowCount && result.rowCount > 0) {
-          console.log(`   \u2705 Table "${table}": Found ${result.rowCount} potential matches. Top score: ${result.rows[0].vector_score.toFixed(4)}`);
+          console.log(`   \u2705 Table "${table}": Found ${result.rowCount} potential matches. Top score: ${result.rows[0].hybrid_score.toFixed(4)}`);
         } else {
           console.log(`   \u26AA Table "${table}": No relevant matches found.`);
         }
         for (const row of result.rows) {
-          const _a = row, { vector_score, id } = _a, rest = __objRest(_a, ["vector_score", "id"]);
+          const _a = row, { hybrid_score, vector_score, keyword_score, id } = _a, rest = __objRest(_a, ["hybrid_score", "vector_score", "keyword_score", "id"]);
           delete rest.embedding;
           const content = `[TYPE: ${table.replace(/s$/, "").toUpperCase()}]
 ` + Object.entries(rest).filter(([k, v]) => v !== null && typeof v !== "object" && k !== "id").map(([k, v]) => `${k}: ${v}`).join("\n");
           allResults.push({
             id: `${table}-${id}`,
-            score: parseFloat(String(vector_score)),
+            score: parseFloat(String(hybrid_score)),
             content,
             metadata: __spreadProps(__spreadValues({}, rest), { source_table: table })
           });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@retrivora-ai/rag-engine",
-  "version": "0.2.3",
+  "version": "0.2.4",
   "description": "Retrivora AI is a plug-and-play AI engine for RAG chat experiences — generic vector DB + LLM provider, embeddable or standalone.",
   "author": "Abhinav Alkuchi",
   "license": "MIT",

package/src/core/Pipeline.ts

@@ -135,7 +135,8 @@ export class Pipeline {
 
     try {
       const queryVector = await this.embeddingProvider.embed(question, { taskType: 'query' });
-      const rawMatches = await this.vectorDB.query(queryVector, topK, ns);
+      // Pass the original text to the vector DB so it can perform hybrid (keyword) search if supported
+      const rawMatches = await this.vectorDB.query(queryVector, topK, ns, { __queryText: question });
 
       const sources = rawMatches.filter((m) => m.score >= scoreThreshold);
       const context = sources.length

package/src/providers/vectordb/MultiTablePostgresProvider.ts

@@ -46,7 +46,7 @@ export class MultiTablePostgresProvider extends BaseVectorProvider {
 
   async initialize(): Promise<void> {
     this.pool = new Pool({ connectionString: this.connectionString });
-    
+
     // Dynamically discover all tables that have an 'embedding' column
     const client: PoolClient = await this.pool.connect();
     try {
@@ -56,9 +56,9 @@ export class MultiTablePostgresProvider extends BaseVectorProvider {
         WHERE column_name = 'embedding' 
         AND table_schema = 'public'
       `);
-      
+
       const discoveredTables = res.rows.map(r => r.table_name);
-      
+
       if (this.tables.length === 0) {
         // No static tables configured, use everything discovered
         this.tables = discoveredTables;
@@ -66,7 +66,7 @@ export class MultiTablePostgresProvider extends BaseVectorProvider {
         // Filter configured tables to ensure they actually exist with embeddings
         const staticTables = [...this.tables];
         this.tables = staticTables.filter(t => discoveredTables.includes(t));
-        
+
         if (this.tables.length < staticTables.length) {
           const missing = staticTables.filter(t => !discoveredTables.includes(t));
           console.warn(`[MultiTablePostgresProvider] Some configured tables were skipped (missing 'embedding' column): ${missing.join(', ')}`);
@@ -123,31 +123,50 @@ export class MultiTablePostgresProvider extends BaseVectorProvider {
 
     console.log(`[MultiTablePostgresProvider] --- Starting Multi-Table Search ---`);
 
+    const queryText = _filter?.__queryText as string | undefined;
+
     for (const table of this.tables) {
       try {
-        // We use a hybrid approach:
-        // 1. Vector similarity (Semantic)
-        // 2. Simple keyword matching (Exact) to boost relevance for specific names
-        const result = await this.pool.query(
-          `SELECT *, 
-           (1 - (embedding <=> $1::vector)) AS vector_score
-           FROM "${table}"
-           ORDER BY vector_score DESC
-           LIMIT 50`,
-          [vectorLiteral]
-        );
+        let sqlQuery = '';
+        let params: any[] = [];
+
+        if (queryText) {
+          // Actual Hybrid Search: Semantic Vector Score + Full-Text Keyword Rank
+          sqlQuery = `
+            SELECT *, 
+             (1 - (embedding <=> $1::vector)) AS vector_score,
+             ts_rank(to_tsvector('english', t.*::text), plainto_tsquery('english', $2)) AS keyword_score,
+             ((1 - (embedding <=> $1::vector)) + (ts_rank(to_tsvector('english', t.*::text), plainto_tsquery('english', $2)) * 2.0)) AS hybrid_score
+            FROM "${table}" t
+            ORDER BY hybrid_score DESC
+            LIMIT 50
+          `;
+          params = [vectorLiteral, queryText];
+        } else {
+          // Fallback to Vector-only Search
+          sqlQuery = `
+            SELECT *, 
+             (1 - (embedding <=> $1::vector)) AS hybrid_score
+            FROM "${table}" t
+            ORDER BY hybrid_score DESC
+            LIMIT 50
+          `;
+          params = [vectorLiteral];
+        }
+
+        const result = await this.pool.query(sqlQuery, params);
 
         if (result.rowCount && result.rowCount > 0) {
-          console.log(`   ✅ Table "${table}": Found ${result.rowCount} potential matches. Top score: ${result.rows[0].vector_score.toFixed(4)}`);
+          console.log(`   ✅ Table "${table}": Found ${result.rowCount} potential matches. Top score: ${result.rows[0].hybrid_score.toFixed(4)}`);
         } else {
           console.log(`   ⚪ Table "${table}": No relevant matches found.`);
         }
 
         for (const row of result.rows) {
-          const { vector_score, id, ...rest } = row as Record<string, unknown>;
+          const { hybrid_score, vector_score, keyword_score, id, ...rest } = row as Record<string, unknown>;
           delete rest.embedding;
 
-          const content = `[TYPE: ${table.replace(/s$/, '').toUpperCase()}]\n` + 
+          const content = `[TYPE: ${table.replace(/s$/, '').toUpperCase()}]\n` +
             Object.entries(rest)
               .filter(([k, v]) => v !== null && typeof v !== 'object' && k !== 'id')
               .map(([k, v]) => `${k}: ${v}`)
@@ -155,7 +174,7 @@ export class MultiTablePostgresProvider extends BaseVectorProvider {
 
           allResults.push({
             id: `${table}-${id}`,
-            score: parseFloat(String(vector_score)),
+            score: parseFloat(String(hybrid_score)),
             content,
             metadata: { ...rest, source_table: table },
           });
@@ -179,7 +198,7 @@ export class MultiTablePostgresProvider extends BaseVectorProvider {
     // 2. Take top 3 from EACH table guaranteed, then fill the rest with overall best
     const balancedResults: VectorMatch[] = [];
     const tables = Object.keys(resultsByTable);
-    
+
     // First, ensure every table's best match is included
     for (const table of tables) {
       balancedResults.push(...resultsByTable[table].slice(0, 3));