Projects

Turning Model Conversations and Agentic Traces into Actionable Insights

A Framework for Efficient and Scalable Neural Retrieval

Advancing the Frontier of Sparse Attention Research

Are Large Reasoning Models Interruptible?

Democratizing Reinforcement Learning for LLMs

AI-Driven Research Systems

Elastic KV Cache for Dynamic GPU Sharing and Efficient Multi-LLM Inference

System Optimization through Reflective Text Evolution

A Least Privilege Framework for Authorizing Tool Calling Agents

A Systematic Evaluation Framework that Automatically Identifies Failure Patterns in LLMs

Reliable and Efficient Semantic Prompt Caching

A Live Benchmark for Generative Research Synthesis

Fast, Accurate, and 100% Private RAG on your Laptop

Retrospective Verification and Self-Correction

Challenging Software Optimization Tasks for Evaluating SWE-Agents

Semantic-Aware Parsing for Security Logs

A Crowdsourced In-The-Wild Evaluation Platform for Search-Augmented LLM Systems Based on Human Preference

Online RL Training for Real-World Long-Horizon Agents

Serving AI Models across Regions and Clouds with Spot Instances

Unlocking Polylogarithmic Accesses in Metadata-Private Messaging

Procedural Environment Generation and Hybrid Verifiers for Scaling Open-Weights SWE Agents

Multi-Agent System Failure Taxonomy

Building Generalist Agents That Scale

A Method for Combining Base Language Models and Instruction-Tuned Language Models for Better Synthetic Data Generation.

A Compositional Framework for Building and Deploying Large Inference-Time Scaling Architectures and Strategies

A Benchmark for Table-Augmented Generation

An Efficient Collective Communication Library for GPUs

Give Your Generative Models a Vibe Check 😀

Next-Generation Open Vision and AI

Train Your Own o1 Preview Model Within $450

LSP client library in python to build applications around language servers

A New Framework for End-to-End Querying of Geospatial Videos

Encrypted Semantic Search with High Accuracy

The Framework for Programming—Not Prompting—Foundation Models

Easily Build Knowledge-Intensive LLM Applications That Reason Over Your Data With LOTUS!

Serving Thousands of Concurrent LoRA Adapters

A Framework for Serving and Evaluating LLM Routers – Save LLM Costs Without Compromising Quality!

Automatic Adapter Selection for Diffusion Models

The First Dynamic Leaderboard for SOTA Text-To-Video Generation Models

Elevating LLM Function Calling with Versatile API Integration

Measuring Function-Calling Capabilities of Different LLMs

A Dynamic Framework for Evaluating AI Coding Systems

A General Solution to Rollback Attacks in TEEs

Automatically Rewriting Distributed Protocols for Scalability

Enabling Replicated State Machines to Communicate Efficiently

Secret Key Recovery in a Global-Scale End-to-End Encryption System

Let Orchestrators Run Your Workloads on Your Cloud Resources Without Handing Over Your Cloud Credentials and Data

A Framework for Deploying On-Demand Distributed Trust

Training Neural Networks on Tiny Devices with Integrated Rematerialization and Paging

A Fast & Accurate Oracle for Object Placement

Holistic and Contamination Free Evaluation of Large Language Models for Code

Blazing Fast Bulk Data Transfers Between Any Cloud

“Retrieval-Augmented Fine-Tuning” combines the benefits of Retrieval-Augmented Generation and Fine-Tuning for better domain adaptation

An Automatic Pipeline to Build High-Quality LLM Benchmarks with High Separability and Agreement to Human Preference from Live Data

A Fast Serving Framework For Large Language Models and Vision Language Models

Building the fastest and easiest-to-use inference engine for LLMs

An Open-Source Chatbot Impressing GPT-4 with 90%* ChatGPT Quality

An Open Platform for Evaluating LLMs by Human Preference

A Runtime for LLM-Generated Actions like Code, API Calls, and More.

A Totally Ordered, High Throughput, Pub/Sub System with Disaggregated Memory