# Albert Tam

![Albert Tam](/resources/cognition/portraits/albert-tam.webp)

![Albert Tam 2](/resources/cognition/portraits/albert-tam.webp)
![Albert Tam 3](/resources/cognition/portraits/albert-tam.webp)

Post-training at OpenAI; ex-Cognition (SWE-grep, SWE-1.5); MIT CS & Math  
OpenAI  
San Francisco Bay Area

## At Cognition

- Role: Former Research Engineer (Engineering)
- Joined: Sep 2025
- How they joined: Directly
- Before Cognition: Algorithm Developer Intern, Hudson River Trading
- Now: OpenAI (https://www.linkedin.com/in/a-s-t)
- Overlapped before Cognition with other Cognition people at: MIT CSAIL; Pika; MIT (https://www.autumn.ai/resources/cognition/worked-together)

## Identity
- LinkedIn: https://www.linkedin.com/in/a-s-t
- X: @albert5tam
- GitHub: https://github.com/tam-albert
- Website: https://tam-albert.github.io/
- Domain: openai.com

## Notes
- Moved to OpenAI as a Member of Technical Staff in Feb 2026 working on post-training, after about six months at Cognition. (https://tam-albert.github.io/)
- 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 :)". (https://x.com/albert5tam/status/2108649244643283349)
- 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. (https://cognition.com/blog/swe-1-5)
- 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. (https://cognition.com/blog/swe-grep)
- 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 :)" (https://x.com/albert5tam/status/1978876860001030320)
- 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. (https://www.linkedin.com/in/a-s-t)
- 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. (https://cognition.com/blog/swe-grep)
- 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. (https://techcrunch.com/2025/07/14/cognition-maker-of-the-ai-coding-agent-devin-acquires-windsurf/)
- Algorithm Developer Intern at Hudson River Trading in New York, May-Aug 2025, doing mid-frequency and high-frequency equities research. (https://www.linkedin.com/in/a-s-t)
- Public headline says "Research at OpenAI". (https://www.linkedin.com/in/a-s-t)
- 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. (https://cognition.com/blog/swe-1-5)
- Research Engineer Intern at Pika (image-to-video models) in Jan 2025. (https://www.linkedin.com/in/a-s-t)
- Research Intern at Phonic in San Francisco, Jun-Aug 2024, on text-to-speech/voice models. (https://www.linkedin.com/in/a-s-t)
- Built zyzx, a natural-language shell in Zig using a Mixtral 8x7B model fine-tuned on bash commands, at TreeHacks 2024. (https://tam-albert.github.io/projects)
- Software Engineer Intern at Inkeep in New York, Jan-Feb 2024, building RAG for documentation search. (https://www.linkedin.com/in/a-s-t)
- Listed on the NanoMath staff team running the NanoMath Fall Math Meet competition alongside Ashwin Rajan, Darren Yao, Johan Ko and Oliver Ni. (https://www.reddit.com/r/math/comments/j6iyqb/upcoming_math_competition_nanomath_fall_math_meet/)
- Led the organizing team for HackMIT (co-director) and Blueprint, MIT's beginner hackathon. (https://tam-albert.github.io/)
- Writes paper notes on his blog, e.g. on DeepSeekMath-V2 (Nov 2025), focusing on verifiers and meta-verifiers for natural-language proof generation. (https://tam-albert.github.io/blog/deepseek-math-v2)
- MIT course projects (Fall 2024): adapting FlatQuant post-training quantization to diffusion transformers (6.5940) and training-free guidance for discrete diffusion language models (6.7960). (https://tam-albert.github.io/projects)
- Undergraduate researcher in the Kellis lab at MIT CSAIL, Sep 2022 - May 2023. (https://www.linkedin.com/in/a-s-t)
- Attended the 2021 USABO national finals (closing convocation June 2, 2021), where the booklet describes him as a junior who enjoys biology, math, computer science, and everything in between. (https://usabo-trc.org/sites/default/files/allfiles/USABO%20Closing%20Ceremony%20Booklet%202021final.pdf)
- Research Intern at Stony Brook University Jun-Dec 2021 on molecular dynamics simulations of hydrogels for soil stabilization (Python, LAMMPS); presented at the MRS 2021 Fall Meeting. (https://www.linkedin.com/in/a-s-t)
- Previous roles include Research Engineer at Cognition; Algorithm Developer Intern at Hudson River Trading; Research Engineer Intern at Pika; Research Intern at Phonic; Researcher at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL); Undergraduate Researcher at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL); Software Engineer Intern at Inkeep; Learning Assistant at Massachusetts Institute of Technology. (https://www.linkedin.com/in/a-s-t)
- Education history includes Bachelor of Science - BS, Computer Science and Mathematics, Massachusetts Institute of Technology. (https://www.linkedin.com/in/a-s-t)
- Extended TRAK to segmentation to curate training subsets up to 3x smaller that improve performance and to flag mislabeled examples. (https://tam-albert.github.io/projects)
- Built Sakhi, a conversational language-learning chatbot, at LAHacks 2023, and StitchIt, a collaborative journaling app, for MIT web.lab 2023. (https://tam-albert.github.io/projects)
- Learning Assistant for MIT 6.190 (Low-Level Programming in C and Assembly), Oct-Dec 2022. (https://www.linkedin.com/in/a-s-t)
- Location is San Francisco Bay Area. (https://www.linkedin.com/in/a-s-t)

## 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 - New York, United States (May 2025 - Aug 2025): Mid-frequency and high-frequency equities research
- Research Engineer Intern at Pika (Jan 2025 - Jan 2025): image-to-video models
- Research Intern at Phonic - San Francisco, California, United States (Jun 2024 - Aug 2024): text-to-speech models
- Researcher at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) - Cambridge, MA (Jun 2023 - May 2024): Data attribution for image segmentation models with Aleksander Madry's lab
- Undergraduate Researcher at MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) - Cambridge, MA (Sep 2022 - May 2023): Kellis lab
- Software Engineer Intern at Inkeep - New York, NY (Jan 2024 - Feb 2024): RAG for documentation search
- Learning Assistant at Massachusetts Institute of Technology - Cambridge, Massachusetts, United States (Oct 2022 - Dec 2022): LA for 6.190 (Low-Level Programming in C and Assembly). Explain CS concepts to students with a wide range of backgrounds.
- Research Intern at Stony Brook University - Stony Brook, New York, United States (Jun 2021 - Dec 2021): Worked with molecular dynamics simulations to engineer hydrogels for sustainable, biodegradable soil stabilization. Used Python and LAMMPS to create visualizations of hydrogel-clay-soil systems. Presented research at the Materials Research Society 2021 Fall Meeting symposium.

## Education
- Massachusetts Institute of Technology - Bachelor of Science - BS, Computer Science and Mathematics - (Sep 2022 - May 2026)

## Sources
- https://www.linkedin.com/in/a-s-t
- https://tam-albert.github.io/
- https://cognition.com/blog/swe-grep
- https://cognition.com/blog/swe-1-5
- https://x.com/albert5tam
- https://x.com/albert5tam/status/2108649244643283349
- https://techcrunch.com/2025/07/14/cognition-maker-of-the-ai-coding-agent-devin-acquires-windsurf/
- https://neurips.cc/virtual/2023/83437
- https://tam-albert.github.io/projects
- https://tam-albert.github.io/blog/deepseek-math-v2
- https://www.cee.org/news-events/press-media/medalists-named-usa-biolympiad-national-finals
- https://usabo-trc.org/sites/default/files/allfiles/USABO%20Closing%20Ceremony%20Booklet%202021final.pdf
- https://www.reddit.com/r/math/comments/j6iyqb/upcoming_math_competition_nanomath_fall_math_meet/
- https://x.com/carlobaronio?lang=en
- https://x.com/tamsbert
- https://github.com/tam-albert
- https://usabo-trc.org/sites/default/files/allfiles/2021%20USABO%20Finalists%20by%20State%20KSF(003
- https://x.com/albert5tam/status/1978876860001030320


Researched with Autumn (https://www.autumn.ai) from public sources. Updated 2026-11-04. Profile: https://www.autumn.ai/resources/cognition/people/albert-tam · Everyone at Cognition: https://www.autumn.ai/resources/cognition/people