Chika Komari

A unified research agenda that integrates structural, semantic, and psychological signal mining.

• Low-dimensional Structural Mining: critical node identification and influence-aware graph mining
• Mid-dimensional Semantic Mining: GraphRAG construction and semantic structure mining for GraphRAG
• High-dimensional Psychological Mining: latent preference discovery and ToM-style preference inference

B.Eng. candidate | GPA 4.04/5.00 | Rank 15/154

• Built a long-term path across complex networks, LLM systems, and preference modeling
• Integrated coursework, research execution, and engineering systems development

AI Infrastructure | Financial Knowledge Graph | Latent Trading Signal Mining

• Designed and implemented an LLM-assisted pipeline for automated financial knowledge graph construction
• Connected graph-serving capabilities to the firm's internal AI assistant and prototyped signal mining workflows
• Improved reproducibility with modular stages and versioned schemas, prompts, and configurations

LLM Reasoning Evaluation | LLM-as-a-Judge

• Contributed to MM-OPERA evaluation design for explanation quality in open-ended association reasoning
• Implemented judge pipelines with GPT-4o and DeepSeek-V3 for scoring, aggregation, and logging
• Supported experiment packaging and internal review artifacts

Computer Vision | Transformer | Architecture Exploration

• Released a transformer-based biomass regression model series on Kaggle Models
• Reached 2.7k+ views and 1.0k+ downloads
• Two related notebooks achieved Gold status

LLM Workflow Orchestration | Data Acquisition | Full-stack Development

• Built multi-turn workflows for detailed QA and resource retrieval
• Implemented authenticated data export/acquisition and a Redis-based producer-consumer pipeline
• Delivered API services and a streaming dashboard with data visualization

Desktop Task Automation | Planner-Executor | Plan Verification

• Converted natural-language requests into Phase-Step-Action execution plans
• Maintained a 43-tool registry with signature-based argument binding
• Added multi-round verification and repair for reliable closed-loop execution

Repo ↗

Personal Knowledge Product | Agent Workflow Orchestration

• Built modules for scholarly search, PDF ingestion, OCR-to-Markdown, and graph-based knowledge visualization
• Supported cross-platform retrieval, topic mining, annotation, and source tracking
• Shipped reproducible setup with Docker Compose

Repo ↗

Integrate structural, semantic, and psychological signal mining into one decision framework, then extend to embodied settings.

• O1-KR1 (by June 30, 2026): critical node identification on multi-layer heterogeneous networks
• O1-KR2 (by September 30, 2026): scalable financial knowledge-graph infrastructure and signal mining
• O2-KR1 (by December 31, 2026): personalized multi-device AI companion based on Airi, benchmarked against Neuro-sama
• Independent KR (by August 2027): secure funding for a personal workstation cluster
• O1-KR3 (by October 31, 2027): transfer critical-node methods from complex networks to semantic networks
• O2-KR2 (by June 30, 2028): full three-direction integration with personalized decision capability

Scholarships, distinctions, research awards, and competitions.

• China National Scholarship (2023-2024), SYSU First-class Scholarship (2023-2024), SYSU Second-class Scholarship (2024-2025)
• Kaggle Notebooks Expert (2025-Present), Best Paper Award (GBCESC 2025)
• Bronze Medal (NeurIPS 2025 Google Code Golf), Provincial First Prize (CUMCM 2025), Meritorious Winner (MCM 2025, Huashu Cup 2025)

Programming, systems engineering, data infrastructure, and LLM tooling.

• Python, TypeScript, C/C++, SQL | FastAPI, Flask, Vue 3, TypeScript
• Neo4j, Redis, SQLite, PostgreSQL, Docker | PyTorch, Hugging Face Transformers
• Dify, Ollama, RAG, Prompt Engineering, LLM Evaluation | Git, Linux, Notion