Products & Features

Core engines and capabilities powering major platforms. Ranked by impact and shipped surface area.

Conversational AI

ChatGPT System

OpenAI

RLHF-tuned conversational agent pipeline serving 100M+ users. Uses PPO and massive cluster orchestration.

LLMRLHF
6 Pubs
RecSys Platform

TikTok Monolith

ByteDance

Real-time recommendation engine with online training and collisionless embedding tables.

RecSysOnline Learning
4 Pubs
Information Retrieval

Google Search Core

Google

The core ranking and retrieval system integrating BERT/MUM and neural matching.

IRNLP
15 Pubs
Model Serving

Llama Inference Stack

Meta

Optimized serving stack for open weights models including quantization and speculative decoding.

InferenceQuantization
3 Pubs
Agent System

Copilot Orchestrator

Microsoft

Grounding engine connecting LLMs to Microsoft 365 graph data and enterprise security boundaries.

RAGAgents
8 Pubs
Compute Engine

CUDA Graph

NVIDIA

Task graph execution engine to reduce CPU launch overhead for high-performance inference.

HPCRuntime
12 Pubs