Jisen Li

About Me

Hello! I'm Jisen Li, an undergraduate student in Mathematics and Computer Science at UIUC, graduating in May 2026. My research interests focus on LLMs for search, ads, and recommendation, as well as multimodal models and agentic AI systems.

In industry, I have interned at Snowflake, working on embedding search models and hybrid retrieval systems, and at TikTok, developing multimodal recommendation models.

In research, I have worked at UChicago and Tsinghua University on AI deployment and computer vision for healthcare. I am now part of the U Lab at UIUC, where I study AI agent communication protocols and build benchmarks for multi-agent systems.

I have also explored entrepreneurship in education, MCN, and international advising—founding and scaling a tutoring agency to 30+ instructors and hundreds of students, and creating tools to help students transfer schools internationally.

Outside of work, I enjoy fitness, billiards, and playing both Go and Texas Hold'em. Feel free to reach out—I’d love to chat!

Interests

• LLM Fine-tuning: Instruction tuning, contrastive learning, CoT, and RLHF for model adaptation.
• Multimodal Learning: Vision-text models and cross-modal alignment using frozen LLMs.
• Search, Ads, and Recommendation: RAG, vector search, and next-item prediction in production.
• Agentic AI: Agents with memory, planning, and decision-making for complex tasks.

Experience

Nexa AI – Incoming Machine Learning Engineer Intern | Oct 2025 – Dec 2025

Model Compression & Distillation: Will work on distilling large language models into lightweight multimodal models for efficient on-device deployment, leveraging quantization (NexaQuant) and hardware-aware optimization.
On-Device ML Infrastructure: Will contribute to nexa-sdk for NPU-based inference, aiming to support cross-platform deployment with a focus on performance and low-latency execution.

Bluelet.ai – Search Tech Lead and Founding MLE | Jun 2025 – Present

Recruiting AI Agent: Autonomously finds, matches, and engages world-class talent—before anyone else does.

Snowflake – Software Engineer Intern (AI/ML) | May 2025 – Aug 2025

Cortex Search: Fine-tuned the Arctic embed v2.0 model for semantic retrieval using contrastive learning, consistency filtering, and CoT-generated queries. Constructed hard negatives with BM25, built an LLM eval framework, and improved nDCG by 15%.
Additionally, developed a Gaussian Process Optimization (GPO) system for auto-tuning hybrid search parameters, optimizing Recall@K and sDCG across retrieval and reranking stages. Final config deployed to Global Service.

TikTok – Machine Learning Engineer Intern | Aug 2024 – Dec 2024

Multimodal Recommendation: Developed Rec_Qwen model based on Qwen2.5-0.5B/7B for multimodal next-item recommendation. Designed a LLaVA-inspired architecture with an MLP projector to align user profiles and content embeddings under a frozen LLM backbone.
Fine-tuned with prompt engineering, improved output structure and loss, and achieved a 28% CTR accuracy gain in real-world ad recommendation settings.

Research

Together.ai | Sep 2025 – Present

KV Cache Quantization: Contributed to building the Channel-wise Precision Boost framework. Supervised by Dr. Shirley Wu.

University of Illinois Urbana-Champaign | Jul 2025 – Present

AI Agent Protocol Benchmarking: Developing a unified benchmark to evaluate AI agent communication protocols across core tasks like Document QA, Collaborative Coding, and MAPF, focusing on task performance, communication cost, and robustness.
Protocol Adaptation and Meta Protocol: Contributed to adapting the ANP protocol across diverse scenarios and integrating a Meta Protocol layer that unifies multiple protocols under a shared interface.
Fail-Storm Recovery and System Monitoring: Implemented failure recovery experiments in the Shard QA scenario and integrated Prometheus + OTLP for monitoring token-level cost, GPU usage, and recovery metrics.
Supervised by Prof. Jiaxuan You.

University of Chicago (Globus Lab) | Jun 2024 – Aug 2024

Automated Deployment for Science Models: Implemented a pipeline to deploy AI/ML for Science models on the Garden platform, integrating the Delta HPC platform for real-time inference. Developed GPU support to enhance Garden's scalability and performance.
Supervised by Dr. Eliu A. Huerta.

Tsinghua University | Jan 2020 – Apr 2021

Non-contact HRV Monitoring: Developed a non-contact heart rate variability (HRV) monitoring algorithm using computer vision and signal processing. Applied OpenCV to extract cheek regions, enhanced subtle color changes via Eulerian Video Magnification (EVM), and designed noise reduction methods for low-light and motion robustness. Extracted HRV features using remote photoplethysmography (rPPG). Improved accuracy by 12% under motion conditions.
Supervised by Prof. Yongqiang Lyu.

Education

University of Illinois Urbana-Champaign (UIUC)
B.S. in Mathematics + Computer Science, GPA: 3.97/4.0
Aug 2023 – May 2026

Huazhong University of Science and Technology (HUST)
B.S. in Computer Science (Transferred)
Sep 2021 – Jan 2023