@retrivora-ai/rag-engine 0.2.8 → 0.3.0

package/dist/server.js

@@ -1528,6 +1528,20 @@ var ConfigValidator = class {
         severity: "error"
       });
     }
+    if (opts.tables && typeof opts.tables !== "string" && !Array.isArray(opts.tables)) {
+      errors.push({
+        field: "vectorDb.options.tables",
+        message: "PostgreSQL tables must be a string or a string array",
+        severity: "error"
+      });
+    }
+    if (opts.searchFields && typeof opts.searchFields !== "string" && !Array.isArray(opts.searchFields)) {
+      errors.push({
+        field: "vectorDb.options.searchFields",
+        message: "PostgreSQL searchFields must be a string or a string array",
+        severity: "error"
+      });
+    }
     return errors;
   }
   /**
@@ -2769,7 +2783,11 @@ var EmbeddingStrategyResolver = class {
 function normalizeHintValue(value) {
   return value.replace(/\s+/g, " ").trim();
 }
+function isLikelyPromptPhrase(value) {
+  return /^(what|which|who|where|when|why|how)\b/i.test(value.trim());
+}
 function extractQueryFieldHints(question) {
+  var _a, _b, _c, _d;
   if (!question.trim()) return [];
   const hints = /* @__PURE__ */ new Map();
   const addHint = (value, field) => {
@@ -2786,6 +2804,46 @@ function extractQueryFieldHints(question) {
   for (const match of question.matchAll(/["']([^"']{2,100})["']/g)) {
     addHint(match[1]);
   }
+  const naturalQuestionPatterns = [
+    /\b(?:what|which)\s+(?:is|are|was|were)\s+(?:the\s+)?([^?.!,]{1,60}?)\s+of\s+["']?([^"'\n?.!,]{2,120})["']?(?=[?.!,]|$)/gi,
+    /\b(?:who|what)\s+(?:is|are|was|were)\s+["']?([^"'\n?.!,]{2,120})["']?(?=[?.!,]|$)/gi,
+    /\b(?:about|for|regarding)\s+["']?([^"'\n?.!,]{2,120})["']?(?=[?.!,]|$)/gi
+  ];
+  const personCompanyPatterns = [
+    /\bcompany(?:\s+name)?\s+(?:of|for)\s+["']?([^"'\n?.!,]{2,120})["']?(?=[?.!,]|$)/gi,
+    /\b(?:which|what)\s+company\s+does\s+["']?([^"'\n?.!,]{2,120})["']?\s+work(?:\s+for|\s+at)?(?=[?.!,]|$)/gi,
+    /\bwhere\s+does\s+["']?([^"'\n?.!,]{2,120})["']?\s+work(?:\s+for|\s+at)?(?=[?.!,]|$)/gi
+  ];
+  for (const pattern of personCompanyPatterns) {
+    for (const match of question.matchAll(pattern)) {
+      const name = match[1];
+      if (name) addHint(name, "name");
+    }
+  }
+  const universalPatterns = [
+    { regex: /([a-zA-Z0-9._%+-]+@[a-zA-Z0-9.-]+\.[a-zA-Z]{2,})/gi, field: "email", group: 1 },
+    { regex: /(\+?\d[\d\-\.\s\(\)]{6,}\d)/g, field: "phone", group: 1 },
+    { regex: /(\b\d{4}-\d{2}-\d{2}\b|\b\d{1,2}\/\d{1,2}\/\d{2,4}\b)/g, field: "date", group: 1 },
+    { regex: /(\$\s?\d{1,3}(?:,\d{3})*(?:\.\d+)?)/g, field: "amount", group: 1 },
+    { regex: /\b(ID|id|identifier)[: ]\s*([A-Za-z0-9\-]{3,})\b/gi, field: "id", group: 2 },
+    // Generic quoted phrase / proper-noun sequences as keywords (already partially handled above)
+    { regex: /"([^"]{2,120})"/g, group: 1 },
+    { regex: /'([^']{2,120})'/g, group: 1 }
+  ];
+  for (const p of universalPatterns) {
+    for (const match of question.matchAll(p.regex)) {
+      const val = p.group ? (_a = match[p.group]) != null ? _a : match[0] : match[0];
+      if (!val) continue;
+      if (p.field) addHint(val, p.field);
+      else addHint(val);
+    }
+  }
+  for (const pattern of naturalQuestionPatterns) {
+    for (const match of question.matchAll(pattern)) {
+      const value = (_b = match[2]) != null ? _b : match[1];
+      if (value) addHint(value);
+    }
+  }
   const fieldPattern = `([^\\n:=?.!,]{1,60}?)`;
   const valuePattern = `([^\\n?.!,]{1,120}?)`;
   const fieldValuePatterns = [
@@ -2795,11 +2853,37 @@ function extractQueryFieldHints(question) {
   ];
   for (const pattern of fieldValuePatterns) {
     for (const match of question.matchAll(pattern)) {
-      addHint(match[2], match[1]);
+      const field = normalizeHintValue((_c = match[1]) != null ? _c : "");
+      const value = (_d = match[2]) != null ? _d : "";
+      if (field && !isLikelyPromptPhrase(field)) {
+        addHint(value, field);
+      } else {
+        addHint(value);
+      }
     }
   }
+  for (const match of question.matchAll(/\b[A-Z][a-z]+(?:\s+[A-Z][a-z]+){0,3}\b/g)) {
+    addHint(match[0]);
+  }
   return [...hints.values()];
 }
+function buildQueryFilter(question, hints) {
+  const filter = { metadata: {}, keywords: [], queryText: question };
+  for (const hint of hints) {
+    if (hint.field) {
+      filter.metadata[hint.field] = hint.value;
+    } else {
+      filter.keywords.push(hint.value);
+    }
+  }
+  for (const match of question.matchAll(/\b[A-Z][a-z]+(?:\s+[A-Z][a-z]+){0,3}\b/g)) {
+    const term = normalizeHintValue(match[0]);
+    if (term && !filter.keywords.includes(term)) filter.keywords.push(term);
+  }
+  if (Object.keys(filter.metadata || {}).length === 0) delete filter.metadata;
+  if (filter.keywords && filter.keywords.length === 0) delete filter.keywords;
+  return filter;
+}
 var Pipeline = class {
   constructor(config) {
     this.initialised = false;
@@ -2888,10 +2972,9 @@ var Pipeline = class {
     try {
       const queryVector = await this.embeddingProvider.embed(question, { taskType: "query" });
       const fieldHints = extractQueryFieldHints(question);
-      const rawMatches = await this.vectorDB.query(queryVector, topK, ns, {
-        __queryText: question,
-        __fieldHints: fieldHints
-      });
+      const filter = buildQueryFilter(question, fieldHints);
+      filter.__entityHints = fieldHints;
+      const rawMatches = await this.vectorDB.query(queryVector, topK, ns, filter);
       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.";
@@ -3744,26 +3827,10 @@ init_PostgreSQLProvider();
 // src/providers/vectordb/MultiTablePostgresProvider.ts
 var import_pg2 = require("pg");
 init_BaseVectorProvider();
-function normalizeFieldName(value) {
-  return value.toLowerCase().replace(/[^a-z0-9]+/g, "");
-}
-function sanitizeFilterHints(rawFilter) {
-  if (!Array.isArray(rawFilter == null ? void 0 : rawFilter.__fieldHints)) return [];
-  return rawFilter.__fieldHints.filter((hint) => typeof hint === "object" && hint !== null).map((hint) => __spreadValues({
-    value: typeof hint.value === "string" ? hint.value.trim().toLowerCase() : ""
-  }, typeof hint.field === "string" && hint.field.trim() ? { field: normalizeFieldName(hint.field) } : {})).filter((hint) => Boolean(hint.value));
-}
-function stripInternalFilterKeys3(filter) {
-  if (!filter) return {};
-  return Object.fromEntries(
-    Object.entries(filter).filter(([key]) => !key.startsWith("__"))
-  );
-}
 var MultiTablePostgresProvider = class extends BaseVectorProvider {
   constructor(config) {
-    var _a, _b, _c;
+    var _a, _b, _c, _d, _e;
     super(config);
-    this.tableSearchConfig = /* @__PURE__ */ new Map();
     const opts = config.options || {};
     if (!opts.connectionString) {
       throw new Error("[MultiTablePostgresProvider] options.connectionString is required");
@@ -3777,6 +3844,8 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
         "[MultiTablePostgresProvider] No tables configured. Set VECTOR_DB_TABLES as a comma-separated list of table names or pass options.tables."
       );
     }
+    const rawSearchFields = (_e = (_d = opts.searchFields) != null ? _d : process.env.VECTOR_DB_SEARCH_FIELDS) != null ? _e : ["name", "product_name", "productname", "title"];
+    this.searchFields = typeof rawSearchFields === "string" ? rawSearchFields.split(",").map((f) => f.trim()).filter(Boolean) : rawSearchFields;
   }
   async initialize() {
     this.pool = new import_pg2.Pool({ connectionString: this.connectionString });
@@ -3802,7 +3871,6 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
       if (this.tables.length === 0) {
         console.warn('[MultiTablePostgresProvider] No tables with "embedding" columns found in the database.');
       } else {
-        this.tableSearchConfig = await this.loadTableSearchConfig(client, this.tables);
         console.log(
           `[MultiTablePostgresProvider] Connected. Searching across ${this.tables.length} table(s): ${this.tables.join(", ")}`
         );
@@ -3832,59 +3900,42 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
    * Query all configured tables and merge results, sorted by cosine similarity score.
    */
   async query(vector, topK, _namespace, _filter) {
-    var _a, _b;
+    var _a, _b, _c;
     if (!this.pool) {
       throw new Error("[MultiTablePostgresProvider] Provider not initialized. Call initialize() first.");
     }
     const vectorLiteral = `[${vector.join(",")}]`;
     const allResults = [];
     console.log(`[MultiTablePostgresProvider] --- Starting Multi-Table Search ---`);
-    const queryText = _filter == null ? void 0 : _filter.__queryText;
-    const fieldHints = sanitizeFilterHints(_filter);
-    const explicitFilter = stripInternalFilterKeys3(_filter);
+    const queryText = _filter == null ? void 0 : _filter.queryText;
+    const entityHints = Array.isArray(_filter == null ? void 0 : _filter.__entityHints) ? _filter.__entityHints.filter(
+      (hint) => typeof hint === "object" && hint !== null && typeof hint.value === "string"
+    ).map((hint) => hint.value.trim().toLowerCase()).filter(Boolean) : [];
     const queryPromises = this.tables.map(async (table) => {
-      var _a2, _b2, _c;
       try {
         let sqlQuery = "";
         let params = [];
-        const tableConfig = this.tableSearchConfig.get(table);
-        const availableFields = (_a2 = tableConfig == null ? void 0 : tableConfig.availableFields) != null ? _a2 : [];
-        const searchableFields = (_b2 = tableConfig == null ? void 0 : tableConfig.searchableFields) != null ? _b2 : [];
-        const normalizedFieldMap = (_c = tableConfig == null ? void 0 : tableConfig.normalizedFieldMap) != null ? _c : /* @__PURE__ */ new Map();
-        const whereConditions = [];
         if (queryText) {
-          const genericValueHints = fieldHints.filter((hint) => !hint.field).map((hint) => hint.value);
-          params = genericValueHints.length > 0 ? [vectorLiteral, queryText, genericValueHints] : [vectorLiteral, queryText];
-          const genericValueParamIndex = genericValueHints.length > 0 ? 3 : -1;
-          const exactValueScoreExpr = genericValueHints.length > 0 && searchableFields.length > 0 ? `GREATEST(${searchableFields.map(
-            (field) => `CASE WHEN LOWER(COALESCE(to_jsonb(t)->>'${field}', '')) = ANY($${genericValueParamIndex}::text[]) THEN 1 ELSE 0 END`
-          ).join(", ")})` : "0";
-          const fieldSpecificChecks = fieldHints.filter((hint) => hint.field).map((hint) => {
-            const column = hint.field ? normalizedFieldMap.get(hint.field) : void 0;
-            if (!column) return null;
-            params.push(hint.value);
-            const paramIndex = params.length;
-            return `CASE WHEN LOWER(COALESCE(to_jsonb(t)->>'${column}', '')) = $${paramIndex}::text THEN 1 ELSE 0 END`;
-          }).filter((expr) => Boolean(expr));
-          const exactFieldScoreExpr = fieldSpecificChecks.length > 0 ? `GREATEST(${fieldSpecificChecks.join(", ")})` : "0";
+          const hasEntityHints = entityHints.length > 0;
+          const exactNameScoreExpr = hasEntityHints ? `+ (
+                 SELECT COALESCE(MAX(
+                   CASE WHEN LOWER(val) IN (${entityHints.map((h) => `'${h.replace(/'/g, "''")}'`).join(", ")})
+                   THEN 1.0 ELSE 0.0 END
+                 ), 0)
+                 FROM jsonb_each_text(to_jsonb(t)) AS kv(key, val)
+                 WHERE key IN (${this.searchFields.map((f) => `'${f}'`).join(", ")})
+               ) * 3.0` : "";
           sqlQuery = `
             SELECT *, 
              (1 - (embedding <=> $1::vector)) AS vector_score,
              COALESCE(ts_rank(to_tsvector('english', t.*::text), NULLIF(REPLACE(plainto_tsquery('english', $2)::text, '&', '|'), '')::tsquery), 0) AS keyword_score,
-             ${exactValueScoreExpr} AS exact_value_score,
-             ${exactFieldScoreExpr} AS exact_field_score,
-             (
-               (1 - (embedding <=> $1::vector)) +
-               (COALESCE(ts_rank(to_tsvector('english', t.*::text), NULLIF(REPLACE(plainto_tsquery('english', $2)::text, '&', '|'), '')::tsquery), 0) * 2.0) +
-               (${exactValueScoreExpr} * 3.0) +
-               (${exactFieldScoreExpr} * 5.0)
-             ) AS hybrid_score
+             ((1 - (embedding <=> $1::vector)) + (COALESCE(ts_rank(to_tsvector('english', t.*::text), NULLIF(REPLACE(plainto_tsquery('english', $2)::text, '&', '|'), '')::tsquery), 0) * 2.0) ${exactNameScoreExpr}) AS hybrid_score
             FROM "${table}" t
             ORDER BY hybrid_score DESC
             LIMIT 50
           `;
+          params = [vectorLiteral, queryText];
         } else {
-          params = [vectorLiteral];
           sqlQuery = `
             SELECT *, 
              (1 - (embedding <=> $1::vector)) AS hybrid_score
@@ -3892,16 +3943,7 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
             ORDER BY hybrid_score DESC
             LIMIT 50
           `;
-        }
-        for (const [key, value] of Object.entries(explicitFilter)) {
-          const column = availableFields.includes(key) ? key : normalizedFieldMap.get(normalizeFieldName(key));
-          if (!column) continue;
-          params.push(String(value));
-          whereConditions.push(`COALESCE(to_jsonb(t)->>'${column}', '') = $${params.length}::text`);
-        }
-        if (whereConditions.length > 0) {
-          sqlQuery = sqlQuery.replace(`FROM "${table}" t`, `FROM "${table}" t
-            WHERE ${whereConditions.join(" AND ")}`);
+          params = [vectorLiteral];
         }
         const result = await this.pool.query(sqlQuery, params);
         if (result.rowCount && result.rowCount > 0) {
@@ -3911,12 +3953,11 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
         }
         const tableResults = [];
         for (const row of result.rows) {
-          const _d = row, { hybrid_score, id } = _d, rest = __objRest(_d, ["hybrid_score", "id"]);
+          const _a2 = row, { hybrid_score, id } = _a2, rest = __objRest(_a2, ["hybrid_score", "id"]);
           delete rest.embedding;
           delete rest.vector_score;
           delete rest.keyword_score;
-          delete rest.exact_value_score;
-          delete rest.exact_field_score;
+          delete rest.exact_name_score;
           const content = `[TYPE: ${table.replace(/s$/, "").toUpperCase()}]
 ` + Object.entries(rest).filter(([k, v]) => v !== null && typeof v !== "object" && k !== "id").map(([k, v]) => `${k}: ${v}`).join("\n");
           tableResults.push({
@@ -3939,12 +3980,23 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
     for (const tableResults of resultsArray) {
       allResults.push(...tableResults);
     }
-    const finalSorted = allResults.sort((a, b) => b.score - a.score);
+    const resultsByTable = {};
+    for (const res of allResults) {
+      const table = (_a = res.metadata) == null ? void 0 : _a.source_table;
+      if (!resultsByTable[table]) resultsByTable[table] = [];
+      resultsByTable[table].push(res);
+    }
+    const balancedResults = [];
+    const tables = Object.keys(resultsByTable);
+    for (const table of tables) {
+      balancedResults.push(...resultsByTable[table].slice(0, 3));
+    }
+    const finalSorted = balancedResults.sort((a, b) => b.score - a.score);
     if (finalSorted.length > 0) {
       console.log(`[MultiTablePostgresProvider] --- Search Complete ---`);
-      console.log(`[MultiTablePostgresProvider] Final top match from "${(_b = (_a = finalSorted[0].metadata) == null ? void 0 : _a.source_table) != null ? _b : "unknown"}" with score ${finalSorted[0].score.toFixed(4)}`);
+      console.log(`[MultiTablePostgresProvider] Final top match from "${(_c = (_b = finalSorted[0].metadata) == null ? void 0 : _b.source_table) != null ? _c : "unknown"}" with score ${finalSorted[0].score.toFixed(4)}`);
     }
-    return finalSorted.slice(0, topK);
+    return finalSorted.slice(0, Math.max(topK, 15));
   }
   async delete(_id, _namespace) {
     console.warn("[MultiTablePostgresProvider] delete() is a no-op for multi-table mode.");
@@ -3965,51 +4017,6 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
       await this.pool.end();
     }
   }
-  async loadTableSearchConfig(client, tables) {
-    var _a, _b;
-    if (tables.length === 0) return /* @__PURE__ */ new Map();
-    const configuredSearchFields = this.parseConfiguredSearchFields();
-    const result = await client.query(`
-      SELECT table_name, column_name, data_type
-      FROM information_schema.columns
-      WHERE table_schema = 'public'
-        AND table_name = ANY($1::text[])
-      ORDER BY ordinal_position
-    `, [tables]);
-    const rowsByTable = /* @__PURE__ */ new Map();
-    for (const row of result.rows) {
-      const rows = (_a = rowsByTable.get(row.table_name)) != null ? _a : [];
-      rows.push({ column_name: row.column_name, data_type: row.data_type });
-      rowsByTable.set(row.table_name, rows);
-    }
-    const configByTable = /* @__PURE__ */ new Map();
-    for (const table of tables) {
-      const columns = (_b = rowsByTable.get(table)) != null ? _b : [];
-      const availableFields = columns.filter(({ column_name }) => column_name !== "embedding").map(({ column_name }) => column_name);
-      const inferredFields = columns.filter(
-        ({ column_name, data_type }) => column_name !== "embedding" && !["ARRAY", "json", "jsonb", "bytea", "tsvector", "USER-DEFINED"].includes(data_type)
-      ).map(({ column_name }) => column_name);
-      const searchableFields = configuredSearchFields.length > 0 ? configuredSearchFields.filter((field) => columns.some((column) => column.column_name === field)) : inferredFields;
-      const normalizedFieldMap = /* @__PURE__ */ new Map();
-      for (const field of availableFields) {
-        normalizedFieldMap.set(normalizeFieldName(field), field);
-      }
-      configByTable.set(table, {
-        availableFields,
-        searchableFields,
-        normalizedFieldMap
-      });
-    }
-    return configByTable;
-  }
-  parseConfiguredSearchFields() {
-    var _a;
-    const raw = (_a = this.config.options) == null ? void 0 : _a.searchFields;
-    if (Array.isArray(raw)) {
-      return raw.filter((field) => typeof field === "string" && field.trim().length > 0);
-    }
-    return [];
-  }
 };
 
 // src/server.ts

package/dist/server.mjs

@@ -34,7 +34,7 @@ import {
   createIngestHandler,
   createUploadHandler,
   getRagConfig
-} from "./chunk-CYVPACA7.mjs";
+} from "./chunk-GT72OIOD.mjs";
 import "./chunk-EDLTMSNY.mjs";
 import {
   PineconeProvider
@@ -335,26 +335,10 @@ function createFromPreset(presetName) {
 
 // src/providers/vectordb/MultiTablePostgresProvider.ts
 import { Pool } from "pg";
-function normalizeFieldName(value) {
-  return value.toLowerCase().replace(/[^a-z0-9]+/g, "");
-}
-function sanitizeFilterHints(rawFilter) {
-  if (!Array.isArray(rawFilter == null ? void 0 : rawFilter.__fieldHints)) return [];
-  return rawFilter.__fieldHints.filter((hint) => typeof hint === "object" && hint !== null).map((hint) => __spreadValues({
-    value: typeof hint.value === "string" ? hint.value.trim().toLowerCase() : ""
-  }, typeof hint.field === "string" && hint.field.trim() ? { field: normalizeFieldName(hint.field) } : {})).filter((hint) => Boolean(hint.value));
-}
-function stripInternalFilterKeys(filter) {
-  if (!filter) return {};
-  return Object.fromEntries(
-    Object.entries(filter).filter(([key]) => !key.startsWith("__"))
-  );
-}
 var MultiTablePostgresProvider = class extends BaseVectorProvider {
   constructor(config) {
-    var _a, _b, _c;
+    var _a, _b, _c, _d, _e;
     super(config);
-    this.tableSearchConfig = /* @__PURE__ */ new Map();
     const opts = config.options || {};
     if (!opts.connectionString) {
       throw new Error("[MultiTablePostgresProvider] options.connectionString is required");
@@ -368,6 +352,8 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
         "[MultiTablePostgresProvider] No tables configured. Set VECTOR_DB_TABLES as a comma-separated list of table names or pass options.tables."
       );
     }
+    const rawSearchFields = (_e = (_d = opts.searchFields) != null ? _d : process.env.VECTOR_DB_SEARCH_FIELDS) != null ? _e : ["name", "product_name", "productname", "title"];
+    this.searchFields = typeof rawSearchFields === "string" ? rawSearchFields.split(",").map((f) => f.trim()).filter(Boolean) : rawSearchFields;
   }
   async initialize() {
     this.pool = new Pool({ connectionString: this.connectionString });
@@ -393,7 +379,6 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
       if (this.tables.length === 0) {
         console.warn('[MultiTablePostgresProvider] No tables with "embedding" columns found in the database.');
       } else {
-        this.tableSearchConfig = await this.loadTableSearchConfig(client, this.tables);
         console.log(
           `[MultiTablePostgresProvider] Connected. Searching across ${this.tables.length} table(s): ${this.tables.join(", ")}`
         );
@@ -423,59 +408,42 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
    * Query all configured tables and merge results, sorted by cosine similarity score.
    */
   async query(vector, topK, _namespace, _filter) {
-    var _a, _b;
+    var _a, _b, _c;
     if (!this.pool) {
       throw new Error("[MultiTablePostgresProvider] Provider not initialized. Call initialize() first.");
     }
     const vectorLiteral = `[${vector.join(",")}]`;
     const allResults = [];
     console.log(`[MultiTablePostgresProvider] --- Starting Multi-Table Search ---`);
-    const queryText = _filter == null ? void 0 : _filter.__queryText;
-    const fieldHints = sanitizeFilterHints(_filter);
-    const explicitFilter = stripInternalFilterKeys(_filter);
+    const queryText = _filter == null ? void 0 : _filter.queryText;
+    const entityHints = Array.isArray(_filter == null ? void 0 : _filter.__entityHints) ? _filter.__entityHints.filter(
+      (hint) => typeof hint === "object" && hint !== null && typeof hint.value === "string"
+    ).map((hint) => hint.value.trim().toLowerCase()).filter(Boolean) : [];
     const queryPromises = this.tables.map(async (table) => {
-      var _a2, _b2, _c;
       try {
         let sqlQuery = "";
         let params = [];
-        const tableConfig = this.tableSearchConfig.get(table);
-        const availableFields = (_a2 = tableConfig == null ? void 0 : tableConfig.availableFields) != null ? _a2 : [];
-        const searchableFields = (_b2 = tableConfig == null ? void 0 : tableConfig.searchableFields) != null ? _b2 : [];
-        const normalizedFieldMap = (_c = tableConfig == null ? void 0 : tableConfig.normalizedFieldMap) != null ? _c : /* @__PURE__ */ new Map();
-        const whereConditions = [];
         if (queryText) {
-          const genericValueHints = fieldHints.filter((hint) => !hint.field).map((hint) => hint.value);
-          params = genericValueHints.length > 0 ? [vectorLiteral, queryText, genericValueHints] : [vectorLiteral, queryText];
-          const genericValueParamIndex = genericValueHints.length > 0 ? 3 : -1;
-          const exactValueScoreExpr = genericValueHints.length > 0 && searchableFields.length > 0 ? `GREATEST(${searchableFields.map(
-            (field) => `CASE WHEN LOWER(COALESCE(to_jsonb(t)->>'${field}', '')) = ANY($${genericValueParamIndex}::text[]) THEN 1 ELSE 0 END`
-          ).join(", ")})` : "0";
-          const fieldSpecificChecks = fieldHints.filter((hint) => hint.field).map((hint) => {
-            const column = hint.field ? normalizedFieldMap.get(hint.field) : void 0;
-            if (!column) return null;
-            params.push(hint.value);
-            const paramIndex = params.length;
-            return `CASE WHEN LOWER(COALESCE(to_jsonb(t)->>'${column}', '')) = $${paramIndex}::text THEN 1 ELSE 0 END`;
-          }).filter((expr) => Boolean(expr));
-          const exactFieldScoreExpr = fieldSpecificChecks.length > 0 ? `GREATEST(${fieldSpecificChecks.join(", ")})` : "0";
+          const hasEntityHints = entityHints.length > 0;
+          const exactNameScoreExpr = hasEntityHints ? `+ (
+                 SELECT COALESCE(MAX(
+                   CASE WHEN LOWER(val) IN (${entityHints.map((h) => `'${h.replace(/'/g, "''")}'`).join(", ")})
+                   THEN 1.0 ELSE 0.0 END
+                 ), 0)
+                 FROM jsonb_each_text(to_jsonb(t)) AS kv(key, val)
+                 WHERE key IN (${this.searchFields.map((f) => `'${f}'`).join(", ")})
+               ) * 3.0` : "";
           sqlQuery = `
             SELECT *, 
              (1 - (embedding <=> $1::vector)) AS vector_score,
              COALESCE(ts_rank(to_tsvector('english', t.*::text), NULLIF(REPLACE(plainto_tsquery('english', $2)::text, '&', '|'), '')::tsquery), 0) AS keyword_score,
-             ${exactValueScoreExpr} AS exact_value_score,
-             ${exactFieldScoreExpr} AS exact_field_score,
-             (
-               (1 - (embedding <=> $1::vector)) +
-               (COALESCE(ts_rank(to_tsvector('english', t.*::text), NULLIF(REPLACE(plainto_tsquery('english', $2)::text, '&', '|'), '')::tsquery), 0) * 2.0) +
-               (${exactValueScoreExpr} * 3.0) +
-               (${exactFieldScoreExpr} * 5.0)
-             ) AS hybrid_score
+             ((1 - (embedding <=> $1::vector)) + (COALESCE(ts_rank(to_tsvector('english', t.*::text), NULLIF(REPLACE(plainto_tsquery('english', $2)::text, '&', '|'), '')::tsquery), 0) * 2.0) ${exactNameScoreExpr}) AS hybrid_score
             FROM "${table}" t
             ORDER BY hybrid_score DESC
             LIMIT 50
           `;
+          params = [vectorLiteral, queryText];
         } else {
-          params = [vectorLiteral];
           sqlQuery = `
             SELECT *, 
              (1 - (embedding <=> $1::vector)) AS hybrid_score
@@ -483,16 +451,7 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
             ORDER BY hybrid_score DESC
             LIMIT 50
           `;
-        }
-        for (const [key, value] of Object.entries(explicitFilter)) {
-          const column = availableFields.includes(key) ? key : normalizedFieldMap.get(normalizeFieldName(key));
-          if (!column) continue;
-          params.push(String(value));
-          whereConditions.push(`COALESCE(to_jsonb(t)->>'${column}', '') = $${params.length}::text`);
-        }
-        if (whereConditions.length > 0) {
-          sqlQuery = sqlQuery.replace(`FROM "${table}" t`, `FROM "${table}" t
-            WHERE ${whereConditions.join(" AND ")}`);
+          params = [vectorLiteral];
         }
         const result = await this.pool.query(sqlQuery, params);
         if (result.rowCount && result.rowCount > 0) {
@@ -502,12 +461,11 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
         }
         const tableResults = [];
         for (const row of result.rows) {
-          const _d = row, { hybrid_score, id } = _d, rest = __objRest(_d, ["hybrid_score", "id"]);
+          const _a2 = row, { hybrid_score, id } = _a2, rest = __objRest(_a2, ["hybrid_score", "id"]);
           delete rest.embedding;
           delete rest.vector_score;
           delete rest.keyword_score;
-          delete rest.exact_value_score;
-          delete rest.exact_field_score;
+          delete rest.exact_name_score;
           const content = `[TYPE: ${table.replace(/s$/, "").toUpperCase()}]
 ` + Object.entries(rest).filter(([k, v]) => v !== null && typeof v !== "object" && k !== "id").map(([k, v]) => `${k}: ${v}`).join("\n");
           tableResults.push({
@@ -530,12 +488,23 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
     for (const tableResults of resultsArray) {
       allResults.push(...tableResults);
     }
-    const finalSorted = allResults.sort((a, b) => b.score - a.score);
+    const resultsByTable = {};
+    for (const res of allResults) {
+      const table = (_a = res.metadata) == null ? void 0 : _a.source_table;
+      if (!resultsByTable[table]) resultsByTable[table] = [];
+      resultsByTable[table].push(res);
+    }
+    const balancedResults = [];
+    const tables = Object.keys(resultsByTable);
+    for (const table of tables) {
+      balancedResults.push(...resultsByTable[table].slice(0, 3));
+    }
+    const finalSorted = balancedResults.sort((a, b) => b.score - a.score);
     if (finalSorted.length > 0) {
       console.log(`[MultiTablePostgresProvider] --- Search Complete ---`);
-      console.log(`[MultiTablePostgresProvider] Final top match from "${(_b = (_a = finalSorted[0].metadata) == null ? void 0 : _a.source_table) != null ? _b : "unknown"}" with score ${finalSorted[0].score.toFixed(4)}`);
+      console.log(`[MultiTablePostgresProvider] Final top match from "${(_c = (_b = finalSorted[0].metadata) == null ? void 0 : _b.source_table) != null ? _c : "unknown"}" with score ${finalSorted[0].score.toFixed(4)}`);
     }
-    return finalSorted.slice(0, topK);
+    return finalSorted.slice(0, Math.max(topK, 15));
   }
   async delete(_id, _namespace) {
     console.warn("[MultiTablePostgresProvider] delete() is a no-op for multi-table mode.");
@@ -556,51 +525,6 @@ var MultiTablePostgresProvider = class extends BaseVectorProvider {
       await this.pool.end();
     }
   }
-  async loadTableSearchConfig(client, tables) {
-    var _a, _b;
-    if (tables.length === 0) return /* @__PURE__ */ new Map();
-    const configuredSearchFields = this.parseConfiguredSearchFields();
-    const result = await client.query(`
-      SELECT table_name, column_name, data_type
-      FROM information_schema.columns
-      WHERE table_schema = 'public'
-        AND table_name = ANY($1::text[])
-      ORDER BY ordinal_position
-    `, [tables]);
-    const rowsByTable = /* @__PURE__ */ new Map();
-    for (const row of result.rows) {
-      const rows = (_a = rowsByTable.get(row.table_name)) != null ? _a : [];
-      rows.push({ column_name: row.column_name, data_type: row.data_type });
-      rowsByTable.set(row.table_name, rows);
-    }
-    const configByTable = /* @__PURE__ */ new Map();
-    for (const table of tables) {
-      const columns = (_b = rowsByTable.get(table)) != null ? _b : [];
-      const availableFields = columns.filter(({ column_name }) => column_name !== "embedding").map(({ column_name }) => column_name);
-      const inferredFields = columns.filter(
-        ({ column_name, data_type }) => column_name !== "embedding" && !["ARRAY", "json", "jsonb", "bytea", "tsvector", "USER-DEFINED"].includes(data_type)
-      ).map(({ column_name }) => column_name);
-      const searchableFields = configuredSearchFields.length > 0 ? configuredSearchFields.filter((field) => columns.some((column) => column.column_name === field)) : inferredFields;
-      const normalizedFieldMap = /* @__PURE__ */ new Map();
-      for (const field of availableFields) {
-        normalizedFieldMap.set(normalizeFieldName(field), field);
-      }
-      configByTable.set(table, {
-        availableFields,
-        searchableFields,
-        normalizedFieldMap
-      });
-    }
-    return configByTable;
-  }
-  parseConfiguredSearchFields() {
-    var _a;
-    const raw = (_a = this.config.options) == null ? void 0 : _a.searchFields;
-    if (Array.isArray(raw)) {
-      return raw.filter((field) => typeof field === "string" && field.trim().length > 0);
-    }
-    return [];
-  }
 };
 export {
   AnthropicProvider,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@retrivora-ai/rag-engine",
-  "version": "0.2.8",
+  "version": "0.3.0",
   "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/config/RagConfig.ts

@@ -31,7 +31,7 @@ export interface VectorDBConfig {
    *
    * For multi-table PostgreSQL search, the following options are also supported:
    * - tables?: string[] | string
-   * - searchFields?: string[] // optional override for which columns receive exact-match boosts
+   * - searchFields?: string[] | string // optional override for which columns receive exact-match boosts
    */
   options: Record<string, unknown>;
 }

package/src/core/ConfigValidator.ts

@@ -151,6 +151,22 @@ export class ConfigValidator {
       });
     }
 
+    if (opts.tables && typeof opts.tables !== 'string' && !Array.isArray(opts.tables)) {
+      errors.push({
+        field: 'vectorDb.options.tables',
+        message: 'PostgreSQL tables must be a string or a string array',
+        severity: 'error',
+      });
+    }
+
+    if (opts.searchFields && typeof opts.searchFields !== 'string' && !Array.isArray(opts.searchFields)) {
+      errors.push({
+        field: 'vectorDb.options.searchFields',
+        message: 'PostgreSQL searchFields must be a string or a string array',
+        severity: 'error',
+      });
+    }
+
     return errors;
   }