Research & Engineering Systems

A temporal-first agentic retrieval system transforming 725,000+ lifelog images into a Neo4j Event Knowledge Graph and Milvus vector store, orchestrated by a time-bound LangGraph agent. Achieved a perfect 1000.0/1000.0 score (Rank #1) on Task 3 (Question Answering) at ACM ICMR LSC 2026.

A reproducible graph machine learning framework that combines analytical first-order Fiedler spectral sensitivity priors with residual GNNs (GCN, GraphSAGE, Edge-MPNN) to estimate road-network connectivity loss across 13 OpenStreetMap areas in 6 Asian countries, achieving a 56× screening speedup on 20,000-node networks.

A two-stage multilingual and cross-lingual claim verification retrieval system evaluated on SemEval-2025 Task 7 across 10 languages. Achieves 0.9163 Overall S@10 and 0.8833 Cross-Lingual S@10 (+8.0% gain from reranking), surpassing the competition winner in cross-lingual retrieval.

A motion-robust contactless vital-sign (BVP & heart rate) estimation framework combining a parameter-free static lower-face anatomical prior with a 3-stream 3D-CNN (SKIA-Net). Achieves 0.7522 Pearson correlation and 3.73 BPM HR MAE on UBFC-rPPG, accepted at ICMV 2026.

A 3-stage multi-view Vietnamese legal document retrieval framework integrating lexical BM25L, Vietnamese dense embeddings, and Graph Attention Networks (GAT) over statutory citation hierarchies with uncertainty-aware meta-learning and decoupled LLM late fusion, achieving 0.9765 Recall@30.

A large-scale biometric privacy evaluation pipeline testing whether super-resolution (Bicubic, SwinIR, ESRGAN) increases identity leakage from degraded contactless finger-selfies across 71,040 probe images and 2,368 livescan templates on ISPFDv2 (Accepted at FDSE 2026, Springer CCIS).

A four-stage conversational AI avatar pipeline (Faster-Whisper ASR → Qwen2.5-7B → Edge-TTS → Wav2Lip) engineered with native in-memory chunk streaming (Stream V2) to eliminate disk I/O bottlenecks, achieving <5s Time-to-First-Frame and >15 FPS video generation.

Official implementation of ViProDID for speaker-disjoint fine-grained Vietnamese dialect identification across all 63 provinces and 3 macro-regions on ViMD, combining `wav2vec2-large-vi` with a 12-dimensional prosodic gating mechanism (59.32% province Macro-F1, 94.42% region Macro-F1).

A constrained expert-routing framework for evidence-grounded fact verification that coordinates a direct verdict expert and a complementary expert via Asymmetric Not-Enough-Information (NEI) Deferral, improving validation Macro-F1 from 0.6920 to 0.7112 (+1.92 points, paired-bootstrap P(Δ>0) = 0.9975).

A modular research software framework for improving fact verification without model parameter updates, featuring immutable provenance-aware prediction caching, data-leakage auditing, fixed-ontology error profiling, and 8 baseline (B0–B8) / 4 retrieval (R1–R4) protocols covered by automated pytest suites.

A real-time, two-stage computer vision pipeline decoupling examinee localization (YOLOv8n) from fine-grained behavioral classification (RexNet-150) across a consolidated 273,897-sample dataset from 10 public sources. Achieves 95.16% test accuracy (+16.64% over full-frame baseline) at 13.9 ms/sample latency (FJCAI 2026).

An experimental research repository for speculative decoding (SPADE) featuring M1 benchmark construction, M2 decoding baseline primitives, and M3 draft-model rejection analysis tools.

An LLM-driven semantic planning and Text-to-SQL/structured query pipeline built for the Zalo AI Challenge Datamate track, translating Vietnamese business analytics questions into schema-grounded, executable relational queries.

A React and TypeScript personal and group expense management web application featuring collaborative budgeting workflows, Firebase persistence, and a Gemini-powered financial assistant.

An interactive Next.js and TypeScript educational web application that visualizes step-by-step Gaussian row elimination and LU matrix factorization (PA = LU) for 2×2 to 5×5 matrices.

A native Android application built with Kotlin, Jetpack Compose, CameraX, and Google ML Kit that performs real-time on-device 33-keypoint human pose detection and skeletal overlay rendering.

A cross-platform HTTP proxy server written in C/C++ featuring raw socket connection handling, HTTP header parsing, configurable domain/IP blacklist filtering, and a lightweight Raylib management GUI.

A from-scratch NumPy implementation and empirical benchmark of 8 search and metaheuristic algorithms (PSO, ABC, Firefly, Cuckoo Search, GA, Hill Climbing, ACO, SA) evaluated across 10D/30D Rastrigin and 20/50-item 0/1 Knapsack problems.