Yinsicheng Jiang
Yinsicheng Jiang is an individual PhD researcher and solo developer/creator, not an organization. His work focuses on algorithm-system co-design and infrastructure for agents and multi-agent systems, with the goal of improving agentic AI workflows for accuracy and efficiency in personal and industrial applications.
At a Glance
- AI and machine-learning systems researchers
- Developers of LLM serving and inference infrastructure
- Teams building agentic and multi-agent applications
- Users of vLLM, SGLang, RAG, Claude Code, and Codex
- +1 more
AI Tools by Yinsicheng Jiang
(1)AgentLedger
Agent Report Review CLI Tool
Discussions
No discussions yet
Be the first to start a discussion about Yinsicheng Jiang
Latest News
AgentLedger repository received a new merge commit; the repository was actively updated under the SecretSettler account.
AgentLedger 0.3.0 added sharing of one report as an .agentledger.zip bundle, safe opening, trace/comment options, and follow-up Ask support for shared traces.
AgentLedger 0.2.0 added same-session Ask, shared mode, one-command setup, a Claude Code plugin marketplace, a Codex skill, and refreshed documentation and visuals.
AgentLedger added privacy protections to keep local checkout paths out of committed store files and traces, and introduced read-only/sandboxed Ask behavior.
Products & Services
An Apache-2.0 open-source CLI and local web interface that files agent reports in a private Git store, renders them for annotation, preserves version history and traces, and routes review questions back to the session that produced the report.
Fast long-context inference via context reuse; the profile reports 4–13× cache hits, 1.5–3× faster prefill, and approximately 36% token savings across vLLM, SGLang, RAG, and AI-agent workloads.
An architecture for scalable and efficient batch inference.
A serverless LLM-inference framework for fast, locality-optimized checkpoint loading and live migration across GPU clusters; the profile says Jiang is a core developer.
Market Position
Jiang's work is positioned as open research infrastructure rather than a conventional company. ContextPilot targets the long-context inference bottleneck alongside serving stacks such as vLLM and SGLang; ServerlessLLM addresses serverless LLM serving; MoE-CAP and TEASBench provide evaluation layers for emerging AI systems; and AgentLedger differentiates from ordinary report viewers or chat-based review by combining Git-backed versioning, agent traces, comments, and session-aware follow-up questions.
Leadership
Founders
Yinsicheng Jiang
PhD student in Computer Science at the University of Edinburgh since January 2025; previously completed a BSc in Computer Science at the University of Edinburgh (September 2020–May 2024). His background includes computer vision for autonomous driving and medical imaging, reinforcement learning, machine-learning systems, 2D-mesh accelerators including Cerebras and Tenstorrent, and a core-developer role on ServerlessLLM.
Founding Story
Jiang's public profile describes a personal research and development practice rather than a separately incorporated venture. He began from machine-learning-systems and computer-vision research and is extending that work toward infrastructure for dynamic agentic and multi-agent workloads, aiming to make such workflows more accurate and efficient.
Business Model
Revenue Model
The publicly documented work is open-source research and software. AgentLedger is distributed under Apache License 2.0 and the README documents installation via GitHub, pipx, or an install script; no paid subscription, API, or commercial pricing model is stated.
Target Markets
- AI and machine-learning systems researchers
- Developers of LLM serving and inference infrastructure
- Teams building agentic and multi-agent applications
- Users of vLLM, SGLang, RAG, Claude Code, and Codex
- Researchers evaluating Mixture-of-Experts, test-time scaling, and agentic workloads
- Research and infrastructure for AI agents and multi-agent systems
- Long-context LLM inference and context-heavy RAG applications
- Agentic workflows requiring efficient repeated inference
- Serverless deployment and multi-GPU LLM serving
- Benchmarking sparse Mixture-of-Experts and emerging AI workloads
- Reviewing, auditing, annotating, and revising AI-generated reports