Arjun Prayoga Aji

Led end-to-end AI product development for one of Indonesias leading securities firms, building and shipping 5 production AI products across customer-facing and internal platforms.

1. Customer Agent AI Chatbot
AI-powered customer service platform with RAG pipeline and real-time human agent suggestion engine. Just launched, targeting 500, 000 active users from a 2.5M account base. Handles 1, 000+ daily messages, reducing response time from ~10 minutes to under 30 seconds.

2. Employee Agent
Internal knowledge AI for policy, compliance, and industry domain Q&A. Multi-agent architecture with graph database (Neo4j) to maintain data hierarchy and retrieval precision. Serving 500+ employees across 11 departments.

3. Conversational Analytics
NL-to-SQL analytics platform enabling self-serve data insights via natural language. Originally scoped for Operations & Research, organically adopted across all 11 departments company-wide.

4. MAIA Trading Assistant Agent
AI assistant built on Miraes proprietary MAIA trading bot, serving 5, 000+ active trading users. Multi-agent architecture with graph database, optimized for retrieval accuracy and speed to support in-app trading
decisions and customer acquisition.

5. E-KYC System
Biometric AI system for customer identity verification integrating face recognition (FaceViT/ViT architecture, OpenVINO-optimized), liveness detection (MobileNet-based), and audio matching (OpenAI Whisper STT).
Just launched, early production rollout underway.

Tech Stack: LangGraph, PostgreSQL, Qdrant, Neo4j, Redis, MinIO, Python, FastAPI, Docker, GCP, OpenVINO, OpenAI Whisper

Principal-level Individual Contributor reporting directly to C-suite, delivering AI products, MIS intelligence, and digital growth across Indonesias largest postal and logistics state-owned enterprise.

1. Led 3-person team to build national Bansos biometric verification system in 1 month under ministerial mandate — replacing 400K/week manual process with automated pipeline processing 60M records in 7 days (150x throughput). YOLO object detection + MobileNet face recognition optimized to 30 FPS at national scale. Reported directly to CEO (FRD).
2. Deployed eKYC system (face recognition + liveness detection) across 5 national products: eBatara Pos, Bansos, PosAja!, PosAja! UMKM, and Dana Pensiun.
3. Architected company-wide MIS data lake integrating 22 servers and 8 departmental data sources, consolidating 60+ executive dashboards across all corporate directorates.
4. Pioneered early RAG + Text-to-SQL system for semantic MIS data retrieval (pgvector, ChromaDB) — enabling natural language queries against operational and sales data, predating LangGraph/LangChain ecosystem maturity.
5. Built early multi-agent system from scratch without dedicated frameworks, demonstrating deep understanding of agentic fundamentals.
6. Drove 15% reduction in reported issues within 3 months and 7%+ MoM MAU growth for PosPay via analysis of 60, 000+ monthly user interactions.
7. Mentored 15+ junior data scientists, conducted monthly workshops, and built data science team from talent to infrastructure.
8. 3x Annual Best Innovator Award for delivering eKYC, RAG Customer Service AI, and Prescriptive Analytics CRM platform.

Stack: YOLO, MobileNet, OpenCV, pgvector, ChromaDB, Python, FastAPI, PostgreSQL, Power BI, Docker

Delivered AI and Data Science curriculum across government programs,
corporate training, and bootcamps — mentoring 25+ students and upskilling 10 corporate transformation talents.

- Kampus Merdeka (2x cohort): Led 20-credit semester program covering Python, ML, Deep Learning, GCP, and Computer Vision (image diffusion, video analysis, image generation).
- Binar Academy: Guided 15 bootcamp students through end-to-end Data Science and NLP curriculum with hands-on project mentorship.
- IFG (Indonesia Financial Group): Corporate trainer for 10 employees on Data Science, ML, Deep Learning, and NLP for company digital transformation initiative.
- Startup Campus: Delivered Computer Vision-focused MSIB curriculum including practical applications in video analysis and image generation.
- Pegadaian: Conducted workshop on LLM and RAG fundamentals for corporate audience.

Topics: Python, SQL, ML, Deep Learning, NLP, Computer Vision, LLM, RAG, GCP, Data Visualization

Built end-to-end AI-powered HR recruitment pipeline, reducing hiring cycle from 3 months to 1 month.
- Engineered OCR + BERT-based document extraction pipeline for automated CV and document parsing, pre-LLM API era.
- Implemented NER to extract structured candidate attributes, fed into 6-dimension NLP scoring model visualized as spider chart for recruiter decision support.
- Improved candidate matching accuracy from 83% to 92%.
- Integrated Amazon Translate for multilingual document handling.

Stack: BERT, NER, Amazon Translate, Python, Deep Learning

• Built AI video analysis system for BCA (COVID-19 monitoring), achieving 90% detection accuracy processing 1, 000+
hours monthly.
• Developed OCR pipeline for PLN smart meter reading using Tesseract transfer learning, reducing manual processing
time by 75%.
• Built KYC identity verification system serving ∼300 customers/month.
Stack: Tesseract, Transfer Learning, OpenCV, Python

Built MLP-based network intrusion detection system for transport layer security, achieving ROC-AUC more than 90% and F1-score = 98% on binary classification (Trojan detection).
Stack: Python, MLP, Scikit-learn