- Joined
- Sep 2025
- How they joined
- Directly
- Before Cognition
- Hudson River TradingAlgorithm Developer Intern
- Now
- OpenAI
Overlapped before Cognition
Details
reinforcement learningpost-trainingcoding agentscontext retrievaldiffusion modelstext-to-speechdata attributionquantitative researchPythonZigRL for coding modelsmath reasoning verifiersmodel efficiency/quantizationhackathons
Notes
- Moved to OpenAI as a Member of Technical Staff in Feb 2026 working on post-training, after about six months at Cognition.Feb 2026
- On 2026-10-09 retweeted Isa Fulford's GPT-6 launch and wrote "intelligence just got a major facelift— and with @AndrewHojel (+many others) we made it way faster as well :)".Oct 9, 2026intelligence just got a major facelift— and with @AndrewHojel (+many others) we made it way faster as well :)
- Credited on Cognition's SWE-1.5 launch (2025-10-29) alongside Jacob Teo, Nikhil Jha, Connor Fogarty, Gary Chang, Theodor Marcu, Edison Zhang and Sean Sullivan; SWE-1.5 is a frontier-size model trained with end-to-end RL in the Cascade agent harness and served on Cerebras at up to 950 tok/s in Windsurf.Oct 29, 2025
- Co-author of Cognition's SWE-grep and SWE-grep-mini blog (2025-10-16) with Ben Pan, Carlo Baronio, Pietro Marsella, Mokshit Jain, Daniel Chiu, Swyx and Silas Alberti: RL-trained fast multi-turn context-retrieval models that power Windsurf's Fast Context subagent.Oct 16, 2025
- Posted on X on SWE-grep launch day, word for word: "ever since we enabled this internally, coding has just felt so much smoother and snappier being a part of this launch has been awesome, excited to share what we've been working on :)"Oct 16, 2025ever since we enabled this internally, coding has just felt so much smoother and snappier being a part of this launch has been awesome, excited to share what we've been working on :)
- Research Engineer at Cognition Sep 2025 - Feb 2026 ("Trained coding models"), joining roughly two months after Cognition signed its agreement to acquire Windsurf (Codeium) in July 2025; SWE-grep and SWE-1.5 both shipped inside the Windsurf product.Sep 2025
- SWE-grep was motivated by Cognition observing that Windsurf and Devin agent trajectories spent >60% of their first turn just retrieving context; the models match frontier coding models at retrieval while taking an order of magnitude less time.Oct 16, 2025
- Cognition acquired Windsurf in July 2025, days after Google DeepMind hired Windsurf's CEO and research leaders; the deal included Windsurf's IP, product, trademark and brand.Jul 14, 2025
- Algorithm Developer Intern at Hudson River Trading in New York, May-Aug 2025, doing mid-frequency and high-frequency equities research.May 2025
- Public headline says "Research at OpenAI".
- SWE-1.5 betas were deployed under the 'Falcon Alpha' name before launch, and the model shipped 13 days after SWE-grep, so Tam was on two back-to-back speed-focused model launches in October 2025.Oct 29, 2025
- Research Engineer Intern at Pika (image-to-video models) in Jan 2025.Jan 2025
Experience
Member of Technical Staff at OpenAI
Feb 2026 - Present
Post-training
Research Engineer at Cognition
Sep 2025 - Feb 2026
Trained coding models
Algorithm Developer Intern at Hudson River Trading
May 2025 - Aug 2025
New York, United States
Mid-frequency and high-frequency equities research
Research Engineer Intern at Pika
Jan 2025 - Jan 2025
image-to-video models
Research Intern at Phonic
Jun 2024 - Aug 2024
San Francisco, California, United States
text-to-speech models
Researcher at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)
Jun 2023 - May 2024
Cambridge, MA
Data attribution for image segmentation models with Aleksander Madry's lab
Undergraduate Researcher at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)
Sep 2022 - May 2023
Cambridge, MA
Kellis lab
Software Engineer Intern at Inkeep
Jan 2024 - Feb 2024
New York, NY
RAG for documentation search
Education
Massachusetts Institute of Technology
Sep 2022 - May 2026
Bachelor of Science - BS, Computer Science and Mathematics
Projects
SWE-grep / SWE-grep-mini (Cognition)
SWE-1.5 (Cognition)
Data Attribution for Segmentation Models (Madry lab)
zyzx: natural-language shell in Zig (TreeHacks 2024)
Sakhi: language-learning chatbot (LAHacks 2023)
StitchIt: collaborative journaling app (web.lab 2023)
FlatQuant for diffusion transformers (MIT 6.5940)
Guidance for diffusion language models (MIT 6.7960)
AchievementsCo-author, SWE-grep / SWE-grep-mini (Cognition, 2025) · Contributor, SWE-1.5 (Cognition, 2025) · NeurIPS 2023 ATTRIB workshop paper: Data Attribution for Segmentation Models · Co-director, HackMIT
VolunteeringHackMIT co-director / organizing lead · Blueprint (HackMIT) organizer · NanoMath staff team
Recent posts
October 2026
intelligence just got a major facelift— and with @AndrewHojel (+many others) we made it way faster as well :)
October 2025
ever since we enabled this internally, coding has just felt so much smoother and snappier being a part of this launch has been awesome, excited to share what we've been working on :)
Articles
Papers
Data Attribution for Segmentation Models (Tam, Vendrow, Madry), NeurIPS 2023 ATTRIB Workshop
Awards2021 USA Biolympiad (USABO) National Bronze Medal · 2021 USABO National Finalist · Two-time USABO semifinalist (top 50, 2020)
Profiles
Code
Languages
PythonHTMLJupyter Notebook
