CV
Curriculum Vitae — Jiantang Huang.
Contact Information
| Name | Jiantang Huang |
| Professional Title | Research Assistant, Human-AI Nexus Group |
| huang.jiant@northeastern.edu | |
| Phone | (339) 221-8946 |
Professional Summary
Research Assistant in Prof. Hang Jiang’s Human-AI Nexus Group, working on LLM agents, agent-to-agent communication, abstract reasoning (ARC-AGI), and world models. MS in Computer Science, Northeastern University.
Experience
-
2026 - Present Boston, MA
Research Assistant
Human-AI Nexus Group (Prof. Hang Jiang)
Working with Prof. Hang Jiang on LLM agents — focusing on agent-to-agent communication and multi-agent coordination.
- Designing protocols and benchmarks for multi-agent LLM coordination.
- Studying agent-to-agent communication patterns in LLM-based systems.
-
2021 - 2022 China
Software Engineer
Huawei Technologies Co., Ltd.
Designed and developed software for the IT application layer, distributed cloud-based platforms, and Internet services using Agile, DevOps, and open-source methodologies.
- Implemented full product lifecycle management — from customer requirements to definition, architecture, development, launch, and operations.
- Solved technical challenges in distributed systems, performance optimization, reliability, and database engineering to strengthen product competitiveness.
- Contributed to security and OSS maintenance work involving Apache and Zlib components.
Education
-
2023 - 2025 Boston, MA, USA
-
2017 - 2021 Columbus, OH, USA
Bachelor of Science
The Ohio State University
Financial Mathematics (minor in Computer Information Science)
Projects
-
Multimodal Traffic Accident Grounding System
Researcher
- Built a zero-shot multimodal accident grounding pipeline that predicts accident time, impact location, and collision type from real CCTV videos using frozen VLMs.
- Designed a two-pass coarse-to-fine framework with Qwen3-VL for temporal-spatial grounding and Gemini 3.1 for type classification, plus confidence-gated fallback mechanisms.
- Integrated YOLO + ByteTrack + physics-based scoring for robust fallback inference and evaluated the system on 2,027 ACCIDENT@CVPR 2026 real CCTV videos.
-
Gemini 3 Global Hackathon — AI Dancing Coach
Developer
- Built an AI-powered dance coaching system that evaluates user performance by comparing dance videos with reference choreography.
- Integrated Gemini to interpret motion differences and generate technique feedback.
- Designed a Retrieval-Augmented Generation (RAG) framework that retrieves dance technique knowledge and choreography guidelines to produce context-aware coaching suggestions.
-
Chain-of-Thought Distillation for Reasoning Enhancement
Developer
- Studied chain-of-thought distillation for reasoning alignment by systematically comparing SFT and preference-based RL (DPO, ORPO) on instruction-tuned LLMs.
- Fine-tuned Flan-T5-Base (250M) on the CoT Collection (1.8M samples) and evaluated on GSM8K and StrategyQA benchmarks.
- Showed that ORPO yields robust reasoning gains under limited compute, improving commonsense accuracy from 14.3% to 40.7% while avoiding PPO cold-start instability.
Publications
-
2026 Two-Pass Zero-Shot Temporal-Spatial Grounding of Rare Traffic Events in Surveillance Video
CVPR Autopilot Workshop, NA Track (Poster); arXiv:2605.01512
Accepted as a poster at the CVPR Autopilot Workshop (NA Track). Coarse-to-fine two-pass VLM pipeline (Qwen3-VL-Plus + Gemini 3.1 Flash-Lite) with confidence-gated fallback. Reaches ACC^S = 0.539 on ACCIDENT@CVPR 2026 (2,027 real CCTV videos), +0.127 over the benchmark paper’s best-of-baselines oracle, at ~$20 total inference cost.
-
2025 Slow-Motion Video Synthesis for Basketball Using Frame Interpolation
arXiv:2511.11644
Frame-interpolation methods for generating high-quality slow-motion basketball footage, with project page at https://biubiugod123.github.io/slow-motion-page/.
Certificates
- Microsoft Certified — Azure AI Fundamentals - Microsoft ()
Skills
Programming Languages (): Python, C / C++, Go, Java, Bash
AI Coding Tools (): Claude Code, Codex, GitHub Copilot
LLMs, Agents & Reasoning (): LLM Fine-tuning (SFT, DPO, ORPO), Multi-agent Systems, Agent-to-Agent Communication, Chain-of-Thought (CoT) Distillation, RAG, Transformers, NLP, World Models, ARC-AGI / Abstract Reasoning, Reinforcement Learning, TensorFlow / PyTorch, MLflow
Computer Vision (): OpenCV, Camera Calibration, Feature Detection (Harris / SURF), 3D Projection, Frame Interpolation
Engineering & Tooling (): Git, CMake, Docker, Microsoft Azure, Unit / Integration Testing, OSS Maintenance (Apache, Zlib)
Languages
English : Professional working proficiency
Chinese (Mandarin) : Native
Interests
Research: Agent-to-agent communication & coordination, Abstract reasoning (ARC-AGI), World models for grounded agents, LLM reasoning & alignment, Multimodal perception