- Joined
- May 2026
- How they joined
- Directly
- Before Cognition
- Jane StreetQuantitative Trader
- Status
- Current
Overlapped before Cognition
“4 months in and 2 models wiser, I can't wait to keep building with this team.”
Details
LocationNew York, New York, United States
Company sitecognition.com
LanguagesChinese (Simplified), English
UniversityMassachusetts Institute of Technology
Python (Programming Language)LeadershipPythonC++JavaLaTeX
Notes
- Credited in the Research byline of Cognition's SWE-1.7 launch (published 2026-07-08), Cognition's most capable model to date, RL post-trained from a Kimi K2.7 base and served in Devin Web, Desktop and CLI via Cerebras at 1000 TPS.Jul 8, 2026
- Joined Cognition as Member of Technical Staff in New York in May 2026 after nearly six years (Aug 2020 - Apr 2026) as a Quantitative Trader at Jane Street; LinkedIn headline now reads "Research @ Cognition".May 2026
- Posted on X on 2026-09-10, word for word: "4 months in and 2 models wiser, I can't wait to keep building with this team." which dates his start to roughly May 2026 and implies he has worked on two model releases.Sep 10, 20264 months in and 2 models wiser, I can't wait to keep building with this team.
- Co-author of FrontierCode 1.1 (2026-07-07) alongside Eric Lu, Ben Pan, Fermi Ma, Alex Lombardi, Deniz Birlikci, Sam Lee, Ray Wang, Rohan Choudhury, Carlo Baronio, Jacob Teo, Joon Hee Lee and Silas Alberti, the eval used as SWE-1.7's headline score.Jul 7, 2026
- Co-author of "Introducing FrontierCode" (2026-06-08), Cognition's benchmark asking whether models write good code, not just correct code; the SWE-1.7 post cites it as reference [1] with TC Qin among ten authors.Jun 8, 2026
- Early-2026 side projects on GitHub, built in the months before leaving Jane Street, form a clear RL ramp: go-bot (Feb 2026, "Builds on Katago to focus on interpretability"), alphago-9x9 (Feb-Mar 2026, "Replicates the original AlphaGo network on a 9x9 board") and cuda-rl (Mar 2026, "Multi-turn RL for writing CUDA kernels").Mar 2026
- In March 2024, while still at Jane Street and the same month Cognition unveiled Devin, he created a public GitHub repo tcqin/devin_ai described as "My attempt at replicating Devin.".Mar 25, 2024
- AlphaStar Academy instructor bio lists: MOP Qualifier (2014); USAJMO (2014: Winner; 2013: Honorable Mention); AMC 10 Perfect Scorer (2014); BMT (2014: 4th Place Individual, 3rd Place Geometry); ARML (2013: 1st Place Team). He taught at the A* 2015 Summer Camp, the same cohort as Sean Shi.2014
- Posted on X on 2026-07-08 (SWE-1.7 launch day), word for word: "Never thought I’d be a part of a team that pushes the pareto frontier for coding!"Jul 8, 2026Never thought I’d be a part of a team that pushes the pareto frontier for coding!
- Mathematics Magazine's report on the 43rd USAMO / 5th USAJMO (2014) lists "Tiancheng Qin 10 Mission San Jose High" among top scorers, matching the USAJMO 2014 winner credential.2014
- Posted on X on 2026-09-08, word for word: "So blessed to be a part of this team", and replied to @cognition the same day with "🚀🚀🚀".Sep 8, 2026So blessed to be a part of this team
- Posted on X on 2026-08-19, word for word: "with devin I never have to choose"; the same day on LinkedIn he wrote, word for word: "At Cognition we never have to choose because Devin AI does it all."Aug 19, 2026At Cognition we never have to choose because Devin AI does it all.
Experience
Member of Technical Staff at Cognition
May 2026 - Present
New York, United States
Quantitative Trader at Jane Street
Aug 2020 - Apr 2026
New York, New York, United States
Software Engineering Intern at Pure Storage
Jun 2017 - Aug 2017
Mountain View, California
I implemented support for SNMPv3 by adding cryptographic security to allow for authentication and privacy encryption. I also modified the GUI and CLI interfaces to support SNMPv3 agents and security measures.
Data Scientist Intern at Twnel
Jan 2017 - Jan 2017
Cambridge, Massachusetts
I developed a recommendation algorithm for the mobile messenger for independent professionals in the emerging marketplace using recurrent neural networks and LSTM models to suggest messaging options.
Education
Massachusetts Institute of Technology
Jan 2016 - Jan 2019
Bachelor’s Degree, Math and Computer Science/Engineering
Coursework: 6.01 (Introduction to EECS) 18.701 (Abstract Algebra) 18.06 (Linear Algebra) 18.03 (Differential Equations)
Massachusetts Institute of Technology
Jan 2016 - Dec 2019
Bachelor’s Degree, Math and Computer Science/Engineering
MIT Poker Club: Committee Member Zeta Beta Tau Syncopasian (MIT's East Asian Premier A Capella Group) Asian Dance Team
Projects
AchievementsResearch credit on Cognition SWE-1.7 (2026-07-08) · Co-author, Introducing FrontierCode (2026-06-08) · Co-author, FrontierCode 1.1 (2026-07-07)
Recent posts
September 2026
4 months in and 2 models wiser, I can't wait to keep building with this team.
@cognition ⬅️ Up and to the left! ⬆️
When I want to factor 260 I use my brain. When I want to factor RSA-260 I talk to @penlume
So blessed to be a part of this team
@cognition 🚀🚀🚀
August 2026
with devin I never have to choose
At Cognition we never have to choose because Devin AI does it all.
July 2026
@wowitsmrinal @cognition Congrats and welcome aboard!
Never thought I’d be a part of a team that pushes the pareto frontier for coding!
Articles
AwardsMOP Qualifier (2014) · USAJMO Winner (2014) · USAJMO Honorable Mention (2013) · AMC 10 Perfect Scorer (2014) · BMT 2014: 4th Place Individual, 3rd Place Geometry · ARML 2013: 1st Place Team
Profiles
Code
Repos9
Repos
Alphago-9x9: Replicates The Original AlphaGo Network On A 9x9 BoardCreate-react-app-lambdaCuda-rl: Multi-turn RL For Writing CUDA KernelsDevin_ai: My Attempt At Replicating Devin.Go-bot: Builds On Katago To Focus On InterpretabilityHeadsup: A NLHE Poker Heads-up Web App.Resy-bot: Bot For Making Reservations On ResyRobot-affect-recognition: Robot Affect Recognition UROP For The Personal Robots Group.
Languages
PythonJavaScript
