Qiran Hu

AC3S: Adaptive Conditioning for 3D-Aware Synthetic Data Generation

• [1] Eric Ji, Qiran Hu, Wufei Ma, Sarthak Jain, Yingying Li, Minh N. Do, Yaoyao Liu. AC3S: Adaptive Conditioning for 3D-Aware Synthetic Data Generation. European Conference on Computer Vision (ECCV) 2026.

Links: arXiv, Code, Video, BibTeX

REVA: Reusable Evidence View Aggregation for Context-Efficient RAG Serving

• [2] Tuan Nguyen, Qiran Hu, Banruo Liu, Khoa D. Doan, Kok-Seng Wong, Fan Lai. REVA: Reusable Evidence View Aggregation for Context-Efficient RAG Serving. IEEE International Conference on Data Mining (ICDM) 2026.

Links: arXiv, Code, BibTeX

ACDN: Agent-Aware Content Delivery Network

• [3] Tuan Nguyen, Qiran Hu, Dibyadeep Saha, Banruo Liu, Khoa D. Doan, Kok-Seng Wong, Fan Lai. ACDN: Agent-Aware Content Delivery Network. Under review.

SV4D 3.0: Single-Step 3D-Aware Diffusion for Multi-View-Consistent 4D Scene Generation

• [4] Qiran Hu, Wei Cao, Yaoyao Liu. SV4D 3.0: Single-Step 3D-Aware Diffusion for Multi-View-Consistent 4D Scene Generation. Under review.

AISim: Using LLM-Simulation as Epistemic Scaffolds for Early Stage Qualitative Research Design

• [5] Hangyue Zhang, Qiran Hu, Ziyi Zhang, Hyanghee Park, Yun Huang. AISim: Using LLM-Simulation as Epistemic Scaffolds for Early Stage Qualitative Research Design. Under review.

AlphaWiSE: Adaptive Weight Interpolation for Continual Multimodal Representation Learning

• [6] Sarthak Jain, Qiran Hu, Zhen Zhu, Yaoyao Liu. AlphaWiSE: Adaptive Weight Interpolation for Continual Multimodal Representation Learning. Under review.

Computer Vision and Machine Learning Group

Undergraduate Research Assistant (Advisor: Prof. Yaoyao Liu), 2025.05 - 2026.08, Champaign, IL

• Proposed adaptive conditioning methods for 3D-aware synthetic data generation to enhance world model understanding and embodied agent performance in interactive simulations with geometry-conditioned diffusion approaches, reducing FID by 32.8% on ImageNet and improving pose accuracy by 4.2x on PASCAL3D+.
• Scaled foundation-model training through distributed methods across TB-level datasets on National Center for Supercomputing Applications (NCSA) HPC clusters to enforce multi-view consistency with 4-bit NF4 quantization and custom CUDA kernels, decreasing FID by 18.3% on AC3S-50 and reducing generation latency by 78.7% at p95.
• Implemented camera-controlled novel view synthesis for video generation pipelines to advance real-time perception for SLAM, visual odometry, and 3D reconstruction, decreasing LPIPS by 21.0% on GSO and FV4D by 52.0% on OmniObject3D compared to SV4D 2.0.

Links: Lab

Multi-modal Generative Models for Continual Learning

Undergraduate Research Assistant (Advisor: Prof. Yaoyao Liu), 2026.02 - 2026.08, Champaign, IL

• Selected for NVIDIA Academic Grant Program Award, AMD University Program, and Lambda Research Grant to streamline class-incremental learning in multimodal foundation models across audio, image, and text without catastrophic forgetting or cross-modal alignment drift, increasing audio-to-text R@1 by 27.6% on AudioSet and raising last-task accuracy by 40.9% on UrbanSound8K.
• Implemented post-hoc tensor-level weight interpolation for multimodal retrieval on National Artificial Intelligence Research Resource (NAIRR) HPC clusters, minimizing trainable parameters from 182M to 499 sigmoid-parameterized coefficients and decreasing training time by 90.2% compared to LwF.
• Optimized checkpoint fusion pipelines to compose frozen checkpoints into one deployable model without additional inference latency, retaining 99.6% of audio-to-text R@1 after shrinking the exemplar memory by 90.0% and enhancing last-task accuracy by 10.9% on CIFAR-100.

Links: NVIDIA Grant

Reusable Evidence View Aggregation for Context-Efficient RAG Serving

Undergraduate Research Assistant, 2026.02 - 2026.06, Champaign, IL

• Designed attention-based scoring for post-retrieval context compression in retrieval-augmented generation (RAG) to reuse generator attention traces across queries and update scores offline, achieving a 91.6% cache hit rate and boosting F1 by 12.9% on HotpotQA.
• Deployed reusable evidence views keyed to unique documents in order to scale compression cost with documents rather than query volume, improving F1 by 13.2% on Natural Questions and decreasing online compression overhead by 6.1x at under 40 ms per request.
• Accelerated online serving to keep context compression off the GPU critical path, cutting compression latency by 99.3% compared to EXIT and 99.8% compared to FaviComp.

Links: IEEE ICDM 2026

Long-Form Video-Language and Audio-Visual Social Understanding

Undergraduate Research Assistant, 2025.12 - 2026.05, Champaign, IL

• Proposed memory-efficient streaming video-language models with temporal transformer blocks for long-form video understanding beyond 30-minute sequences, boosting zero-shot accuracy by 17.4% and cutting peak memory by 58.2% compared to full-sequence attention.
• Developed multimodal fusion pipelines to combine facial action units, body pose, and prosody through cross-modal attention and infer social intent and conversational role from raw recordings, achieving 89.1% accuracy on speaker-role classification over a 2,108-turn bootstrap.
• Scaled end-to-end annotation pipelines for full-length clinical sessions with automatic transcription and labeling, processing recordings 1.4x to 1.7x faster than real time and increasing speaker-role inversion accuracy to 90.0%.

Multi-agent Research Synthesis Engine

Undergraduate Research Assistant, 2025.11 - 2026.05, Champaign, IL

• Orchestrated eight specialized agents across a 12-step workflow for literature synthesis to cover planning, retrieval, drafting, reflection, and safety review in a single loop with LLM-as-judge evaluation, scoring 0.955 overall and 0.925 on relevance, safety, and clarity.
• Developed concurrent multi-source retrieval across Semantic Scholar and Tavily with production-grade fallback handling for API failures, cutting query latency from 8.2 s to 4.9 s, a 40.2% reduction.
• Designed customized Model Context Protocol servers to centralize governed tool access behind unified interfaces over academic databases, code repositories, and document stores, cutting serving latency by 87.5%.

Links: Project Page

Realistic Neural Style Transfer Architecture

Undergraduate Research Assistant, 2025.01 - 2025.08, Champaign, IL

• Proposed style transfer frameworks to maintain photorealistic results under highly abstract styles with VGG perceptual losses and edge-preserving constraints, improving SSIM by 77.0% and MS-SSIM by 49.0% compared to TensorFlow's NST.
• Optimized multi-layer Gram matrix losses with adaptive layer weighting to suppress texture and chromatic artifacts, boosting SSIM by 4.4x and MS-SSIM by 6.4x compared to ChatGPT-4o.
• Formulated Sobel edge and total variation losses to keep contours sharp and suppress high-frequency noise, decreasing LPIPS by 23.0% and DISTS by 17.0% compared to TensorFlow's NST.

Links: GitHub

Anime Statistics and Analysis Platform, ASAP

Undergraduate Research Assistant, 2025.02 - 2025.06, Champaign, IL

• Deployed interactive R Shiny analytics platforms to surface anime popularity trends and market opportunities with live Jikan REST API data on 18,000 titles and predictive analysis on shinyapps.io.
• Developed R data pipelines to fetch, clean, and store 18,000 anime titles from the Jikan REST API in SQLite, adding genre, studio, and producer fields for five interactive analysis and recommendation modules.
• Trained random forest models to predict anime scores from engagement features, decreasing test RMSE by 90.1% and raising R² by 3.7% compared to linear regression.

Links: GitHub

IBM Research

Research Intern (Advisor: Dr. Dhaval Patel), 2026.09 - Present, New York, NY

• Lead Model Context Protocol (MCP) server architecture for AssetOpsBench that serves failure-mode knowledge graphs, root-cause ranking, and intervention estimation to multi-turn tool-use large language model (LLM) agents via deterministic Neo4j queries with provenance, raising top-1 tool accuracy by 12.9% compared to per-server sub-agents and cutting standing tool context by 98.5%.
• Design fault-diagnosis reinforcement learning (RL) environments on a hydraulic test rig with verifiable cost-aware rewards, priced observations, and reward-hacking checks for multi-step LLM agents, raising top-1 diagnosis accuracy by 42.6 percentage points compared to a Plan-Execute agent and eliminating 87.7% of model calls per episode.
• Streamline the diagnosis environment into standard Gymnasium and OpenEnv training interfaces for RLVR post-training with GRPO on verl, minimizing per-episode inference latency by 52.7% across 47 recorded fault combinations and saving 47.5% in maintenance cost compared to a do-nothing policy.

Links: AssetOpsBench

Memoria

Technical Lead, 2026.01 - 2026.08, Champaign, IL

• Led end-to-end agentic workflow deployments across MCP servers, sub-agents, and agent skills with multi-agent orchestration, permission-aware tool calling, and policy guardrails, accelerating product delivery by 4.0x and boosting overall task success by 17.9% on CL-bench.
• Managed production operations for agentic workflows with logging, tracing, evaluation harnesses, and monthly iteration cycles for client stakeholders, reducing serving cost by 90.5% compared to GPT-4o and improving rule-application task success by 31.2% on CL-bench.
• Integrated recursive self-evolving memory architecture across sessions with cost-aware model routing into enterprise agent workflows, saving 96.0% in token cost compared to GPT-5.1 and increasing procedural task success by 71.4% on CL-bench.

Links: Website

Two by Two Learning Center

Full Stack Developer, 2025.08 - 2025.12, Champaign, IL

• Launched NOODEIA, a multi-agent tutoring platform for K-12 students below grade level, to plan, critique, and monitor each user with long-horizon memory, boosting user confidence by 2.44x and increasing independence by 2.35x in counterbalanced within-subjects studies.
• Implemented recursive self-evolving long-term memory on Neo4j to retrieve contextually relevant interactions, accelerating query speed by 3.6x compared to recency-biased FIFO memory on PostgreSQL and minimizing cross-session retrieval to 2.1 s.
• Deployed complexity-aware model selection across planner, retrieval, solver, and critic stages to score each request and route it to the smallest capable model, eliminating 89.9% of monthly serving cost compared to GPT-4o and routing 75% of calls to lighter model tiers.

Links: Website

WRC

Data Analyst, 2024.08 - 2024.12, Champaign, IL

• Built paired pre/post analytics pipelines over survey data from 9,935 incoming students, raising correct responses by 14.0% and decreasing ambiguous responses by 61.9%.
• Conducted A/B tests on two consent scenarios to locate where misconceptions persisted after each workshop, increasing accuracy on consent comprehension by 19.5% and cutting persistent misconceptions by 41.6%.
• Designed scenario-based learning modules for underrepresented subgroups, raising correct responses on bystander scenarios by 36.8% and narrowing the gap in correct-response rates between subgroups 5.6x.

CS 107 Data Science Discovery

Teaching Assistant, 2023.08 - 2026.05, Champaign, IL

• Led weekly lab sections and office hours for in-person and online sessions, mentoring 1,200 students every semester and 7,200+ over six semesters for data science foundations in Python, statistical inference, data wrangling, and machine learning.
• Authored DISCOVERY Guides on the course website to provide detailed explanations with applied Python and statistics walkthroughs, giving students a self-serve reference for every lab and homework.
• Designed problem sets, exam questions, test suites, and autograder scripts to deliver instant, consistent feedback to 1,200 students every semester on Mastery Platform, cutting grading turnaround from days to minutes.

Links: Guides, Mastery

OnePromptClaudeCode

Lead Developer, 2026.03 - Present, Champaign, IL

• Maintain an MIT-licensed open-source agent harness for agentic software development, packaging 95 skills, 71 slash commands, 14 specialized agents, 4 hooks, and 10 MCP servers into a one-prompt install that sets up the full plan, build, review, test, and ship workflow on a fresh machine, eliminating 99.5% of environment initialization steps.
• Automate capability routing with a three-tier router that matches each prompt to the skills and agents it needs by keyword, then by session-memory recall, and only then by a 4.5 s model reasoning pass, to load only the matched capabilities and optimize tokens, latency, and cost on every turn, resolving keyword-matched prompts in 120 ms and decreasing routing latency by 97.3% compared to model-only routing.
• Improve the security layer with four lifecycle hooks that screen every tool call before execution, auto-deny destructive shell commands, and confine edits to a directory-scoped sandbox, cutting accidental destructive actions by 32.6%.

Links: GitHub

Technical Skills

• Programming Languages and AI/ML Frameworks: Python, C++, C, Rust, Go, Java, Swift, Kotlin, Ruby, R, PyTorch, CUDA, JAX, TensorFlow, Triton, TensorRT, vLLM, SGLang, NeMo, Megatron-LM, LangGraph, NCCL, GPU/TPU/CPU architecture.
• Foundation Model Training: continual learning, synthetic data generation, distributed training, quantization, context compression, knowledge distillation, pre-training, post-training, reinforcement learning, RLHF, reward modeling, supervised fine-tuning (SFT), direct preference optimization (DPO), PPO, GRPO, RLAIF, RLVF, model alignment and instruction following, test-time training, token pruning, kernel optimization, LoRA, QLoRA, FSDP.
• Computer Vision and Agentic AI: diffusion models, 3D/4D image and video generation, novel view synthesis, multi-view geometry consistency, 3D reconstruction, world models, depth estimation, visual-inertial odometry, SLAM, spatial intelligence, autoregressive models, flow matching, NeRFs, 3D Gaussian Splatting, OpenCV, multi-agent orchestration, tool calling, agent harness, long-horizon agent memory, context engineering, RAG, MCP, policy guardrails, sub-agent parallelization, prompt caching, computer-use agents, A2A.
• Full-stack, Databases, Infrastructure, and Design: TypeScript, JavaScript, React, Tailwind CSS, FastAPI, HTML5, Vue, Angular, PostgreSQL, Neo4j, MongoDB, Kafka, Docker, Kubernetes, CI/CD, cloud platforms (AWS, GCP, Azure), Figma, Canva, Adobe Creative Suite, Microsoft Office Suite, Unity.
• Languages: Chinese (native), English (native), Spanish (limited working proficiency).

Certifications and Honors

• Neo4j Certified Professional
• Neo4j Graph Data Science Certification

Links: Neo4j Professional, Neo4j GDS

Education

• Columbia University, New York, NY
  M.S. in Data Science
  Fu Foundation School of Engineering and Applied Science
  Courses: High Performance Machine Learning, Algorithms, Continual Learning and Memory Models, Computer Systems
  2026.09 - Expected 2028.05
• University of Illinois Urbana-Champaign, Champaign, IL
  B.S. in Data Science and Information Science
  Minors: Computer Science and Statistics
  Siebel School of Computing and Data Science
  Honors: Dean's List and James Scholar
  Courses: Applied Machine Learning, Generative AI for Human-AI Collaboration, Advanced AI Web-App Development, Graph Databases, Data Visualization, Computational Photography, Linear Algebra with Computational Applications
  2022.08 - 2026.05

Contact

• Email: qh2332@columbia.edu
• Phone: +1 (347) 957-9176

Links: GitHub, LinkedIn, X, Website, Google Scholar