Technical Highlights
Multi-Agent AI System: Architected dan implemented 10 specialized AI agents dengan LangGraph orchestration, handling complex workflows dari database queries hingga automated model retraining.
Production ML Pipeline: Complete MLOps setup dengan XGBoost model (95.45% recall), automated retraining, drift detection, dan model versioning menggunakan MLflow + DagSHub.
Full-Stack Development: Led development sebagai ketua kelompok - backend (FastAPI, PostgreSQL), AI chatbot (18K+ lines), dan partial frontend (React chat UI).
Enterprise-Grade: JWT authentication, role-based access, comprehensive error handling, monitoring, dan deployment ke Google Cloud Run dengan Docker.
Skills Demonstrated
AI/ML Engineering
- ▸Multi-Agent Systems: LangGraph state machine dengan 10 specialized agents
- ▸LLM Integration: AWS Bedrock (Claude 3.5 Sonnet) dengan streaming responses
- ▸Vector Database: Qdrant untuk knowledge base (SOPs, manuals, FAQs)
- ▸Natural Language to SQL: LLM-based query generation dengan temporal context parsing
- ▸Prompt Engineering: Complex prompts untuk agent collaboration dan reasoning transparency
Machine Learning & MLOps
- ▸XGBoost: 44 engineered features, hyperparameter tuning, 95.45% recall
- ▸Feature Engineering: Rolling averages, rate of change, interaction terms
- ▸Model Versioning: MLflow + DagSHub untuk experiment tracking
- ▸Automated Retraining: Trigger-based retraining (performance degradation, data drift)
- ▸Drift Detection: KS-test untuk detect distribution changes
- ▸Model Monitoring: Performance metrics tracking, feature importance analysis
Backend Development
- ▸FastAPI: 2,000+ lines REST API dengan async endpoints
- ▸PostgreSQL: Database design dengan SQLAlchemy 2.0, performance indexes
- ▸Authentication: JWT-based auth dengan Argon2 password hashing
- ▸API Design: RESTful endpoints untuk dashboard, machines, tickets, simulation
- ▸Error Handling: Comprehensive error handling dengan retry logic
AI Chatbot Development
- ▸LangGraph Orchestration: 10,000+ lines multi-agent system
- ▸State Management: Complex state machine dengan conditional routing
- ▸Agent Collaboration: Inter-agent communication protocol
- ▸Bilingual Support: Indonesian/English dengan context-aware detection
- ▸Reasoning Transparency: Show AI thinking process untuk build trust
Frontend Development
- ▸React 19: Modern UI dengan TypeScript
- ▸Real-time Updates: WebSocket integration untuk live data
- ▸Chat Interface: Built chat UI component untuk AI copilot
- ▸State Management: React hooks untuk complex state
DevOps & Infrastructure
- ▸Docker: Multi-stage builds untuk optimized images
- ▸Google Cloud Run: Serverless deployment dengan auto-scaling
- ▸CI/CD: GitHub Actions untuk automated testing dan deployment
- ▸Monitoring: Logging, metrics, alerts setup
- ▸Database Migrations: Alembic untuk version-controlled schema changes
Team Leadership
- ▸Project Management: Led 5-person team, sprint planning, code reviews
- ▸Technical Decisions: Architecture design, tech stack selection
- ▸Documentation: Comprehensive docs untuk onboarding dan maintenance
- ▸Mentoring: Pair programming, knowledge sharing sessions
The 10 AI Agents
Agent A (qdrant_search): Knowledge base search untuk SOPs dan manuals dengan semantic search.
Agent B (database_query): Natural language to SQL dengan temporal context parsing ("hari ini", "minggu lalu").
Agent C (predictive_maintenance): ML prediction dengan XGBoost (95.45% recall, 87.2% precision).
Agent D (web_search): Latest information dari internet menggunakan Tavily API.
Agent E (optimization_engine): Schedule optimization dengan priority scoring, budget constraints, technician availability.
Agent F (simulation_engine): What-if analysis untuk delay impact (cost increase, risk level).
Agent G (feedback_loop_analyzer): Model performance monitoring dengan drift detection (KS-test).
Agent H (intelligent_retrainer): Automated model retraining dengan comparison dan deployment.
Agent I (report_generator): PDF report generation dengan charts dan insights.
Agent J (ticket_creator): Natural language ticket creation dengan multi-turn conversation.
ML Model Performance
XGBoost V3:
- ▸Recall: 95.45% (catch 95 dari 100 failures)
- ▸Precision: 87.2%
- ▸F1-Score: 91.2%
- ▸ROC-AUC: 96.8%
Why High Recall? Dalam predictive maintenance, missing a failure bisa catastrophic (safety risk, $260K/hour downtime). Better have false positives than miss real failures.
Features: 44 engineered features including:
- ▸Raw sensor values (temperature, RPM, torque, vibration, pressure)
- ▸Rolling averages (3, 7, 14 days)
- ▸Rate of change
- ▸Interaction terms (temp × RPM)
- ▸Statistical features (std, min, max)
Architecture Complexity
Codebase: 18,000+ lines of production code
- ▸backend.py: ~2,000 lines (REST API, database, auth)
- ▸main.py: ~10,000 lines (AI chatbot, 10 agents, LangGraph)
- ▸app.py: ~200 lines (unified entry point)
Database: 8 tables dengan complex relationships
- ▸users, authentication, machine_sensor_data
- ▸scheduled_maintenance, machine_data_backup
- ▸chat_thread, chat_message, simulation_history
API Endpoints: 30+ endpoints across 6 modules
- ▸Authentication, Dashboard, Machines, Tickets, Prioritization, Simulation
Technical Challenges Solved
Multi-Agent Coordination: 10 agents harus collaborate tanpa conflict. Solution: LangGraph state machine dengan conditional routing dan agent collaboration protocol.
State Management: Complex state across agents dengan retry logic. Solution: TypedDict state schema dengan proper error handling.
Cost Optimization: AWS Bedrock per-token pricing bisa mahal. Solution: Cache language detection, optimize prompt length, monitor usage dengan alerts.
Bilingual Support: Indonesian/English dengan context-aware detection. Solution: LLM-based detection dengan conversation history context.
Production Deployment: Environment management, database migrations, monitoring. Solution: Docker containerization, CI/CD pipeline, comprehensive logging.
Model Retraining: Automated retraining tanpa downtime. Solution: Blue-green deployment strategy dengan model comparison sebelum switch.
Business Impact
For Engineers:
- ▸Keputusan lebih cepat (analyze sensor data dari hours ke minutes)
- ▸Proactive maintenance (predict failures 3-7 days ahead)
- ▸Natural language interface (no need SQL atau complex tools)
For Business:
- ▸Reduce unplanned downtime (early detection dengan 95.45% recall)
- ▸Cost savings (optimize maintenance scheduling)
- ▸Improved safety (catch critical failures early)
- ▸Data-driven decisions (replace gut feeling dengan predictions)
ROI: Dengan 95.45% recall dan $260K/hour downtime cost, system bisa save millions annually.
Leadership & Collaboration
Role: Ketua kelompok di ASAH program (Dicoding × Accenture)
Responsibilities:
- ▸Full backend development (FastAPI, PostgreSQL, authentication)
- ▸AI chatbot development (10 agents, LangGraph orchestration)
- ▸ML pipeline (XGBoost training, MLOps setup)
- ▸Partial frontend (chat UI component)
- ▸Team coordination (sprint planning, code review, deployment)
Team Size: 5 members dengan different skill levels
Duration: 5 months intensive (900+ hours learning + capstone)
Read Full Story: Blog Post