feat: implement AI-based matching engine with scoring and inference c… by dientruonggg · Pull Request #23 · dscdut/D-Shiftify

dientruonggg · 2026-05-20T02:39:11Z

This pull request introduces a new AI-powered job matching feature to the backend, leveraging vector search (pgvector), dynamic scoring, and Google Gemini AI for job-profile fit explanations. The implementation includes new API endpoints, database migrations, AI logic, and comprehensive tests. It also adds new dependencies to support AI and validation workflows.

AI Matching Feature Implementation

API & Routing
- Added a new REST endpoint GET /api/ai/match/:profileId that returns a list of jobs matched to a user profile, using the new AI matching logic. (ai.controller.js, ai.resolver.js, index.js) [1] [2] [3]
AI Matching Logic
- Implemented matchJobsForProfile agent: fetches the user profile and CV, performs vector similarity search on jobs using pgvector, dynamically infers scoring weights for each job using Google Gemini AI, calculates a hybrid score, and returns the top matches with AI-generated explanations. (match.agent.js)
- Added weight inference logic using Google Gemini AI, with fallback to defaults and caching. Weights are validated with Zod and persisted to the database. (match.inference.js)
- Implemented scoring functions for skills, experience, accessibility, geography, and hybrid score aggregation. (match.scoring.js)

Database & Dependencies

Database Migration
- Added a migration to enable the vector extension, and add embedding and weights_json columns to jobs and narrative_embedding to profiles, supporting vector search and AI metadata. (20260424130019_add_pgvector_and_ai_fields.js)
Dependencies
- Added @google/genai for Gemini API, zod for schema validation, jest and babel-jest for testing. (package.json) [1] [2] [3]
- Updated .gitignore to exclude yarn.lock.

Testing

Comprehensive Unit Tests
- Added tests for the agent orchestrator, inference logic, and scoring logic, ensuring correct error handling, weight inference, and scoring calculations. (match.agent.test.js, match.inference.test.js, match.scoring.test.js) [1] [2] [3]

References:
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14]

…apabilities, and perform comprehensive codebase refactoring and dependency updates.

vercel · 2026-05-20T02:39:16Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
d-shiftify	Ready	Preview, Comment	May 20, 2026 3:05am

Copilot

Pull request overview

Adds an AI-assisted job matching capability to the backend, combining vector similarity (pgvector) with rule-based scoring and optional Gemini-based inference/explanations, plus an HTTP endpoint to fetch matches for a profile.

Changes:

Introduces matching engine components (scoring, weight inference, orchestration agent) and Jest tests for them.
Adds DB migration for pgvector and new embedding/weights fields.
Exposes a new /ai/match/:profileId API route and adds new dependencies (Gemini SDK + Zod) and test tooling.

Reviewed changes

Copilot reviewed 12 out of 16 changed files in this pull request and generated 11 comments.

Show a summary per file

File	Description
package-lock.json	Adds a root lockfile (currently minimal/empty).
backend/src/core/modules/ai/agents/match/match.scoring.js	Implements geo/accessibility/skill/experience scoring and a hybrid score combiner.
backend/src/core/modules/ai/agents/match/match.inference.js	Infers and caches per-job scoring weights via Gemini; persists weights to DB.
backend/src/core/modules/ai/agents/match/match.agent.js	Orchestrates profile/job retrieval, vector search, scoring, and explanation generation.
backend/src/core/modules/ai/agents/match/tests/match.scoring.test.js	Unit tests for scoring functions.
backend/src/core/modules/ai/agents/match/tests/match.inference.test.js	Unit tests for weight inference behavior and fallbacks.
backend/src/core/modules/ai/agents/match/tests/match.agent.test.js	Unit tests for the orchestrator/agent behavior with mocks.
backend/src/core/database/migrations/20260424130019_add_pgvector_and_ai_fields.js	Adds pgvector extension + embedding/weights columns.
backend/src/core/api/index.js	Registers the new AI module in the API module resolver list.
backend/src/core/api/ai/ai.resolver.js	Adds an AI route resolver (currently not aligned with the repo’s Module system).
backend/src/core/api/ai/ai.controller.js	Adds controller to serve matching results.
backend/package.json	Adds Jest test script and new dependencies (`@google/genai`, `zod`, `jest`, `babel-jest`).
backend/.gitignore	Ignores `yarn.lock` (and normalizes Dockerfile entry).
.gitignore	Adds root ignores including `node_modules` and `backend/yarn.lock`.

Comments suppressed due to low confidence (1)

backend/src/core/database/migrations/20260424130019_add_pgvector_and_ai_fields.js:21

exports.down chains .raw('DROP EXTENSION...') off a schema builder, which is not a supported Knex pattern in this codebase (and likely won’t execute). Use sequential await knex.schema.alterTable(...) calls, then await knex.raw('DROP EXTENSION IF EXISTS vector;').

exports.down = function(knex) {
  return knex.schema
    .alterTable('profiles', table => {
      table.dropColumn('narrative_embedding');
    })
    .alterTable('jobs', table => {
      table.dropColumn('embedding');
      table.dropColumn('weights_json');
    })
    .raw('DROP EXTENSION IF EXISTS vector;');

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-05-20T03:05:41Z

+exports.up = function(knex) {
+  return knex.schema.raw('CREATE EXTENSION IF NOT EXISTS vector;')
+    .alterTable('jobs', table => {
+      table.specificType('embedding', 'vector(768)');
+      table.jsonb('weights_json');
+    })
+    .alterTable('profiles', table => {
+      table.specificType('narrative_embedding', 'vector(768)');
+    });
+};
+
+exports.down = function(knex) {
+  return knex.schema
+    .alterTable('profiles', table => {
+      table.dropColumn('narrative_embedding');
+    })
+    .alterTable('jobs', table => {
+      table.dropColumn('embedding');
+      table.dropColumn('weights_json');
+    })
+    .raw('DROP EXTENSION IF EXISTS vector;');


Fixed in 7d6696a. I split the migration into sequential awaited steps (knex.raw + separate knex.schema.alterTable calls) in both up and down.

Copilot · 2026-05-20T03:05:44Z

+import { Router } from 'express';
+import { AIController } from './ai.controller.js';
+
+export const AIResolver = {
+  /**
+   * @param {Router} router
+   */
+  resolve: (router) => {
+    const aiRouter = Router();
+
+    /**
+     * @swagger
+     * /ai/match/{profileId}:
+     *   get:
+     *     tags: [AI]
+     *     summary: Lấy danh sách việc làm phù hợp với profile
+     *     parameters:
+     *       - in: path
+     *         name: profileId
+     *         required: true
+     *         schema:
+     *           type: string
+     *     responses:
+     *       200:
+     *         description: Thành công
+     */
+    aiRouter.get('/match/:profileId', AIController.matchForProfile);
+
+    router.use('/ai', aiRouter);
+  }
+};


Fixed in 7d6696a. AIResolver now uses the Module.builder().addPrefix(...).register([...]) pattern instead of a plain object resolver.

Copilot · 2026-05-20T03:05:46Z

+export const AIController = {
+  /**
+   * GET /api/ai/match/:profileId
+   */
+  matchForProfile: async (req, res, next) => {
+    try {
+      const { profileId } = req.params;
+      const matches = await matchJobsForProfile(profileId);
+
+      return res.status(200).json({
+        success: true,
+        data: matches,
+        message: 'Tìm việc làm phù hợp thành công.'
+      });
+    } catch (error) {
+      next(error);


Fixed in 7d6696a. AIController.matchForProfile now follows project conventions (accepts req only and returns ValidHttpResponse.toOkResponse(...)).

Copilot · 2026-05-20T03:05:49Z

+const deg2rad = (deg) => deg * (Math.PI / 180);
+
+export const geoScore = (lat1, lon1, lat2, lon2) => {
+  if (!lat1 || !lon1 || !lat2 || !lon2) return 0;


Fixed in 7d6696a. geoScore now uses explicit null/undefined checks and numeric validation, so 0 coordinates are treated as valid values.

Copilot · 2026-05-20T03:05:52Z

+export const inferWeights = async (job) => {
+  if (weightCache.has(job.id)) return weightCache.get(job.id);
+
+  if (job.weights_json) {
+    const w = typeof job.weights_json === 'string' ? JSON.parse(job.weights_json) : job.weights_json;
+    weightCache.set(job.id, w);
+    return w;
+  }


Fixed in 7d6696a. weights_json parsing is now guarded, validated with Zod, normalized, and safely falls back to defaults when invalid.

Copilot · 2026-05-20T03:05:57Z

+Dockerfile
+
+yarn.lock


Fixed in 7d6696a. I removed yarn.lock from .gitignore so the existing committed lockfile remains tracked.

Copilot · 2026-05-20T03:06:00Z

+  const profile = {
+    ...profileRecord,
+    skills: cvRecord.skills || [],
+    experiences: cvRecord.experiences || []
+  };


Fixed in 7d6696a. I now normalize/load additional profile data from CV JSON (conditions) including accessibility_needs and optional coordinates before scoring.

Copilot · 2026-05-20T03:06:03Z

+export const atMatchScore = (needs = [], job) => {
+  let s = 0;
+  const env = (job.work_environment || '').toLowerCase();
+
+  if (needs.includes('screen_reader')) {
+    if (['nvda', 'jaws', 'screen_reader'].some((k) => env.includes(k))) s += 0.5;
+    else if (job.accessibility_level === 'AAA') s += 0.3;
+    else if (job.accessibility_level === 'AA')  s += 0.15;
+  }
+  if (needs.includes('voice_control') && job.is_remote) s += 0.3;
+  if (job.accessibility_level === 'AAA') s += 0.2;
+  else if (job.accessibility_level === 'AA') s += 0.1;
+


Fixed in 7d6696a. atMatchScore now maps accessibility scoring to existing schema data (work_mode) while preserving existing compatibility checks.

Copilot · 2026-05-20T03:06:06Z

+export const skillScore = (candidate_skills = [], job_skills = [], semantic_score = 0) => {
+  if (!job_skills || !job_skills.length) return semantic_score;
+  const jLower = job_skills.map((s) => s.toLowerCase());
+  const exact = candidate_skills.filter((s) => jLower.includes(s.toLowerCase())).length;
+  return Math.min(exact / job_skills.length + semantic_score * 0.3, 1);


Fixed in 7d6696a. skillScore now normalizes both string and object skill shapes (e.g. { name: ... }) before matching.

Copilot · 2026-05-20T03:06:08Z

+  const cvRecord = await knex('cvs').where({ profile_id: profileId }).first() || {};
+
+  const profile = {
+    ...profileRecord,
+    skills: cvRecord.skills || [],
+    experiences: cvRecord.experiences || []
+  };


Fixed in 7d6696a. In match.agent I now parse/normalize JSONB-backed CV/job fields (skills, experiences, conditions) before scoring and explanation generation.

Agent-Logs-Url: https://github.com/dscdut/D-Shiftify/sessions/9dbcbe07-c588-4bf2-bad3-afc62c3e0db1 Co-authored-by: dientruonggg <109971402+dientruonggg@users.noreply.github.com>

feat: implement AI-based matching engine with scoring and inference c…

4330389

…apabilities, and perform comprehensive codebase refactoring and dependency updates.

dientruonggg requested a review from Copilot May 20, 2026 02:39

Copilot started reviewing on behalf of dientruonggg May 20, 2026 02:39 View session

dientruonggg marked this pull request as ready for review May 20, 2026 02:43

Copilot AI reviewed May 20, 2026

View reviewed changes

dientruonggg requested review from Anh-Thuw, PhamVanTien1910, ThuongHoai081, hodinhtuankiet and kietoichoiDXD May 20, 2026 02:51

dientruonggg marked this pull request as draft May 20, 2026 02:58

dientruonggg removed request for Anh-Thuw, PhamVanTien1910, ThuongHoai081, hodinhtuankiet and kietoichoiDXD May 20, 2026 02:58

Copilot started work on behalf of dientruonggg May 20, 2026 02:59 View session

fix: address PR review feedback for AI matching flow

7d6696a

Agent-Logs-Url: https://github.com/dscdut/D-Shiftify/sessions/9dbcbe07-c588-4bf2-bad3-afc62c3e0db1 Co-authored-by: dientruonggg <109971402+dientruonggg@users.noreply.github.com>

vercel Bot deployed to Preview May 20, 2026 03:05 View deployment

Copilot finished work on behalf of dientruonggg May 20, 2026 03:08

dientruonggg marked this pull request as ready for review May 22, 2026 16:25

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat: implement AI-based matching engine with scoring and inference c…#23

feat: implement AI-based matching engine with scoring and inference c…#23
dientruonggg wants to merge 2 commits into
devfrom
feat/DS22

dientruonggg commented May 20, 2026 •

edited

Loading

Uh oh!

vercel Bot commented May 20, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Copilot AI May 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		Dockerfile

		yarn.lock No newline at end of file

Uh oh!

Conversation

dientruonggg commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vercel Bot commented May 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI May 20, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

dientruonggg commented May 20, 2026 •

edited

Loading

vercel Bot commented May 20, 2026 •

edited

Loading