Devano JonathanAI Engineer & Data Scientist.
Transforming raw data into strategic impact through AI automation. Building autonomous agents with human-level reasoning.
AI Agent Development
VLM + LLM
Building autonomous testing agents at Hitachi Rail using Vision-Language Models.
Business Impact
100+ hrs saved
Automated HR screening for 3,000+ candidates per cycle at Career Hackers.
ML & NLP
30% accuracy↑
Fine-tuned SBERT models for semantic similarity and cheating detection.
About Me
Data Science and Engineering student at The University of Hong Kong focused on transforming raw data into strategic impact through AI automation. Currently building autonomous agents capable of navigating complex professional tools with human-level reasoning. Expert in bridging the gap between technical insights and business execution.
Education
The University of Hong Kong
Bachelor of Engineering in Data Science and Engineering
Sep 2022 - Dec 2026 (Expected)
Relevant Coursework:
Impact Lab
Social Impact Leadership
Jan 2024 - May 2024
SMA Ignatius Global School
Natural Science
Jul 2019 - May 2022
Key Achievements
Bloomberg Global Trading Challenge 2024
2nd Runner-up in Asia-Pacific. Demonstrated advanced quantitative analysis and risk management strategies in a competitive trading environment against top finance students globally.
Saudi Arabia Algo Trading Challenge 2025
Global Finalist. Developed an algorithm that generated +21% return over 2 months using statistical arbitrage and advanced quantitative strategies.
Indonesia Maju Scholarship
Awarded full scholarship from Indonesian Ministry of Education via Informatics Olympiad National Finals qualification, recognizing academic excellence and competitive programming prowess.
Work Experience
7 internships across AI, Data Science, and Software Engineering.
Software Placement Intern – Machine Learning
- ▹Built autonomous VLM testing agent using Quantized Vision-Language Models on vLLM for UI automation
- ▹Deployed REST API for real-time client-server communication enabling loop-testing without human intervention
- ▹Integrated EasyOCR + OpenCV hybrid pipeline for screenshot verification against backend logs
- ▹Automated VM provisioning via Python, creating scalable CI/CD for multi-server parallel testing
Summer Intern, Full Stack GenAI Developer – AI Research Team
- ▹Built automated cheating detection system using speech-to-text for online interviews
- ▹Improved detection accuracy by 30% via fine-tuned Sentence-BERT with cosine similarity
- ▹Automated screening for 3,000+ candidates/cycle, saving ~100 hours of HR workload per batch
Summer Intern, Research Analyst
- ▹Architected market intelligence taxonomy categorizing 1,200+ tech companies
- ▹Led end-to-end data pipeline development achieving 99.8% accuracy
- ▹Delivered 20+ sector analysis reports to Tier-1 investment banks
Technical Expertise
Comprehensive toolkit for complex challenges
Python
ExpertC++
AdvancedR
IntermediateMATLAB
IntermediateBash
AdvancedTensorFlow
AdvancedScikit-learn
ExpertHugging Face
AdvancedOpenCV
AdvancedEasyOCR
AdvancedVLM
AdvancedLLMs (SBERT)
ExpertRAG Architecture
ExpertAgentic AI
ExpertNLP
AdvancedPandas
ExpertPySpark
AdvancedNumPy
ExpertSeaborn
AdvancedMatplotlib
ExpertTableau
AdvancedGoogle Looker
AdvancedFinancial Modeling
AdvancedQuant Analysis
AdvancedSQL
ExpertMySQL
AdvancedMongoDB
IntermediateRPA
AdvancedREST APIs
ExpertDocker
AdvancedGit/GitHub
ExpertJupyter
ExpertExcel (VBA)
AdvancedPivotTables
ExpertReact
AdvancedNext.js
AdvancedNode.js
AdvancedTailwind CSS
AdvancedFeatured Projects
Selected work in AI, Data Science, and Web Dev
Neuro-Symbolic Geometric AI Agent
Hybrid AI Architecture fusing LLMs (ChatGPT) with geometric engines (GeoGebra) to solve complex problems with >95% verified accuracy. Developed a Validation API that converts natural language into geometric constructions with alternative reasoning paths. Fine-tuning with Reinforcement Learning (PPO, GRPO, DAPO) to enforce logical consistency.
LLMsGeoGebraPPOGRPORLAPIAutonomous VLM Testing Agent
Built at Hitachi Rail: High-throughput visual testing agent using Quantized VLMs on vLLM for low-latency inference. Autonomously interprets UI screenshots, converting visual semantics into executable X/Y coordinates. Integrated EasyOCR + OpenCV for hybrid vision pipeline.
VLMvLLMOpenCVEasyOCRREST APIRPA
Leadership & Community
Building communities and driving data science literacy
HKU Data Science Association
FounderThe University of Hong Kong
Community Building
Established a thriving community of 50+ members passionate about data science, machine learning, and AI applications.
Strategic Planning & Execution
Managed strategic planning to promote data science literacy across the university, organizing workshops, hackathons, and industry networking events.
Blog & Insights
Coming Soon
I'm working on sharing insights about AI, machine learning, data science, and finance. Stay tuned for deep-dives into technical concepts, project breakdowns, and industry trends.