You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit fd5120b
Browse filesBrowse the repository at this point in the historyBrowse files
Restructure impact section; add talks; clean up dead search code
- Split impact paragraph into DeepMind and Bespoke subsections with proper lists and source links
- Add BlackRock and Snowflake panels to recent talks
- Remove dead /content.json search handler and orphan Cmd+K binding
- Tighten ol li spacing via scoped .impact-list class
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
<h2class="section-title">Recent Work & Impact</h2>
49
+
<divclass="row align-items-center">
50
50
<divclass="col-lg-8">
51
51
<p>
52
52
I started <ahref="https://www.bespokelabs.ai">Bespoke Labs</a> in 2024. We focus on RL environment curation and data curation.
53
53
</p>
54
-
<p>
55
-
Previously, I was a Tech Lead in Google DeepMind and worked on LLM-based recommender models. In 2021,
56
-
very few people were thinking about applying the bitter lesson to recommender systems.
57
-
I kickstarted a project to scale recommender models using LLMs, which resulted in
58
-
several research publications (and one best paper award) and many impactful launches across recommender systems such as YouTube.
59
-
</p>
54
+
55
+
<h3>Impact from Google DeepMind</h3>
56
+
<olclass="impact-list">
57
+
<li>First to productionize TPUs for recommender systems at YouTube — reshaped YouTube's compute strategy.</li>
58
+
<li>Demonstrated the gains of scaling recommender models — reshaped YouTube's modeling strategy.</li>
59
+
<li>Introduced <ahref="https://arxiv.org/abs/2305.05065">Semantic IDs and Generative Retrieval for Recommender Systems</a>, now deployed or researched across YouTube, LinkedIn, Snapchat, Alibaba, <ahref="https://research.atspotify.com/2025/9/semantic-ids-for-generative-search-and-recommendation">Spotify</a>, Meta, and <ahref="https://arxiv.org/abs/2506.13695">Kuaishou</a>. This has completely changed the course of research being done in the recommender systems community because it laid down the foundation for using generative AI for recommender models.</li>
60
+
<li>Over the course of my tenure at Google, I had more than 50 peer and spot bonues, 5+ perfy awards, 1 best paper award, 1 feats of engineering award.</li>
61
+
</ol>
62
+
63
+
<h3>Impact from Bespoke Labs</h3>
64
+
<olclass="impact-list">
65
+
<li>Instrumental in getting <ahref="https://openthoughts.ai">OpenThoughts</a> started (<ahref="https://x.com/madiator/status/1884284103354376283">origin thread</a>).</li>
66
+
<li>Core contributors to <ahref="https://arxiv.org/pdf/2601.11868">Terminal Bench</a>, the most popular agentic benchmark.</li>
0 commit comments