AICIN: AI Course Intelligence Network
A production-grade multi-agent recommendation system delivering personalized AI course matches in under 3 seconds using 6 specialized Cloud Run microservices.
The Challenge: Finding Signal in 4,000 Courses
LearningAI365 hosts an extensive catalog of nearly 4,000 AI courses organized into 251 distinct learning paths. The challenge? Matching learners with their ideal educational journey from this vast sea of options.
We needed an intelligent system that could handle real production data, deliver truly personalized recommendations, scale automatically with demand, and reduce costs compared to our previous AWS Lambda implementation.
The solution: AICIN—a sophisticated multi-agent recommendation engine built entirely on Google Cloud Run infrastructure.
3,950
AI Courses
251
Learning Paths
2-3s
Response Time
100%
Success Rate
Six Specialized Agents, One Unified Architecture
AICIN decomposes the recommendation challenge into six intelligent agents, each running as an independent Cloud Run container with auto-scaling from zero to 100 instances based on real-time demand.
01
Orchestrator
Coordinates all agents with JWT authentication and request routing
02
Profile Analyzer
Converts quiz responses into structured learner profiles in 200ms
03
Content Matcher
Applies TF-IDF NLP for semantic similarity across 251 paths (800ms)
04
Path Optimizer
Executes 2-layer scoring: 30% content match, 70% metadata fit (600ms)
05
Course Validator
Ensures complete, high-quality data in all recommendations
06
Recommendation Builder
Formats explainable, actionable results for end users
The Optimized Quiz Experience
We streamlined the assessment from 15 questions down to 9—a 40% reduction—while maintaining recommendation quality through data-driven analysis.
The quiz captures seven critical dimensions: experience level, specific interests, weekly time availability, budget constraints, desired timeline, certification requirements, and learning style preferences.
Clean interface design with progress tracking keeps users engaged through completion, setting the stage for personalized recommendations.
Experience Level
Beginner through Advanced
Interest Areas
ML, Python, Computer Vision, NLP
Time & Budget
10-20 hrs/week, $100-$500
Learning Goals
Career upskilling, hands-on projects
2.5 Seconds to Personalized Intelligence
In just 2.5 seconds with warm instances, AICIN analyzes learner profiles against 251 learning paths using sophisticated 7-dimensional weighted scoring.
78% Match
Complete Computer Vision Journey
Perfect alignment with intermediate ML focus and project-based learning preference
77% Match
Google Cloud Vision API Path
Strong fit for Python expertise and cloud-native implementation experience
74% Match
Intermediate Computer Vision
Excellent balance of technical depth and practical application
Notice the differentiation: these aren't arbitrary scores. TF-IDF semantic analysis and multi-dimensional scoring produce meaningful distinctions that reflect true learner-path compatibility.
Distributed Processing in Action
Parallel Agent Execution
When you submit the quiz, the Orchestrator coordinates all six agents in an optimized workflow. Profile Analyzer and Content Matcher execute in parallel for maximum efficiency.
1
Profile Extraction
200ms - Quiz data structured into learner profile
2
TF-IDF Analysis
800ms - Semantic matching across 251 paths
3
7D Scoring
600ms - Weighted optimization computation
4
Validation & Build
400ms - Quality checks and formatting
Performance Characteristics
2-3s
Warm instance response time
14s
Cold start initialization
251
Paths analyzed per request
After the first request initializes all agents, Cloud Run keeps instances warm for fast, consistent performance on subsequent requests.
Proven Results: 100% Success Rate
Rigorous testing validates AICIN's production readiness across diverse learner profiles and use cases.
Computer Vision Intermediate
Correctly matched CV interests with appropriate intermediate-level paths
Machine Learning Beginner
Identified foundational ML paths suitable for newcomers to the field
Advanced NLP Specialist
Recommended sophisticated NLP tracks matching expert-level requirements
3/3
Tests Passed
2.2s
Avg Response
78-51%
Score Range
Excellent score differentiation from 78% for optimal matches down to 51% for lower-priority paths demonstrates sophisticated analytical capabilities—not just returning uniform scores.
Why Google Cloud Run Powers AICIN
Auto-Scaling to Zero
Pay nothing during idle periods. Instances spin up in seconds during peak demand. Each service scales independently—Content Matcher can reach 100 instances while Profile Analyzer stays at 5.
Deep Integration
Vertex AI Gemini 1.5 Flash enriches recommendations with natural language insights. Memorystore Redis caches frequent queries. Secret Manager secures credentials. Cloud Logging provides full observability.
Resilient Architecture
Circuit breaker patterns prevent cascade failures. Each microservice operates independently, ensuring system reliability even if individual agents experience issues.
AICIN: Multi-Agent Architecture, Production-Ready Performance
Complete source code, architecture documentation, and comprehensive test results available on GitHub.
Built for the Google Cloud Run Hackathon. Powered by Cloud Run, Vertex AI, and Memorystore.
