— Acheul Studio · Open 2024 · Bay Area

Modeling
human psyches.

Acheul Studio investigates how human psyches can be modeled — and, increasingly, how they might be represented inside machines. It is the current vehicle for a question its founder, Taiga Sawamura, has pursued for more than fifteen years: one that has run through financial-market behavior, generative media, and now the internal mechanisms of large language models.

The studio runs small-scale experiments, works with industry and academic collaborators on a project basis, and brings ideas to market only when they prove substantially useful.

— Projects Six works · 2024 – Ongoing
P.01 2026 — Ongoing

Representing Affect in Language Models

How can affect — the shifting dynamics of human emotional states — be represented inside a machine? This project brings computational neuroscience and psychology together with mechanistic interpretability, using activation steering to study how affect can be represented in transformer models.

Pursued as a research direction through campus-wide collaboration at Stanford.

Mechanistic interpretability · Computational neuroscience & psychology

P.02 2026 — Ongoing

Generative AI and the Future of Content

Once affect dynamics run inside multi-modal models, content can be generated at scale and at speed — with depth and breadth, and personalized to the individual. The open question is what people will actually engage with when that becomes possible, and why. This thread studies that question through audience response.

Pursued through prototyping and large-scale user surveys at Stanford GSB.

Generative media · Audience modeling · Survey research

P.03 2025 Past

Real-time Generative Media Models

A prototype that rewrites live video through generative models at interactive frame rates — off-the-shelf image models aggressively modified for real-time inference. The work opened a direction in generative media, and a sharper question: how audiences actually respond to synthetic content.

Developed in close collaboration with the co-founding team.

Model training · Evaluation · Infrastructure

P.04 2025 Past

All Your Apps, Enabled for Glasses

People won't accept AR glasses that lock them out of the apps they already live in. So enabling every smartphone application — rendered AR-native rather than ported as flat rectangles — wasn't optional; it was the requirement. We built it, and concluded that it's meaningful but lacks the gravity of a smartphone screen.

Developed in close collaboration with the co-founding team.

OS engineering · Systems · UX

P.05 2025 Past

Fashionable AR Glasses

Waveguide optics has changed the AR landscape — lower cost, lighter weight, finally near consumer-ready. The remaining question is whether the experience inside the glass is compelling enough to justify wearing them at all.

Acheul worked extensively with the Asian supply chain to find a viable path from prototype to mass production, alongside full-time co-founders who joined from Google's AR division (Tech Lead) and the Fitbit / Google hardware team (Staff Mechanical Engineer).

Hardware, AR

P.06 2024 Past

Ultralightweight VR Eyewear

Nobody wants to wear a brick over their eyes — not even at 100 grams. The idea was to radically narrow the experience the device needed to deliver, then radically reduce the sensors and components required to deliver it. Under 100 grams remains, with today's components, an open problem.

Explored down to every component, with engineers who previously led work at Apple, Meta, Google, NVIDIA, Fitbit, and Magic Leap.

Hardware, VR

— About

Taiga Sawamura.

Taiga Sawamura was born in Japan and raised in North Carolina and New York. He is based in the San Francisco Bay Area, where he is pursuing research on affect and language models at Stanford.

Taiga has spent more than 15 years modeling human psyches in various domains, including investor and market sentiment, borrower–lender game theory, and generative media consumption.

Taiga co-founded Emerada, a Japanese fintech startup that runs behind over 1,000 banks and one million small and medium enterprises. As CEO, he helped grow the company to a $100M valuation, raising $20M to date from institutional investors including JAFCO (often dubbed Japan's Sequoia). He led development of one of the first credit-risk models driven by real-time cash-flow data; that work contributed to academic research on Bayesian hierarchical credit scoring published in the International Journal of Financial Engineering (2021). Before Emerada, he advised M&A and financing at Goldman Sachs.

After selling his stakes in Emerada, Taiga founded Acheul Studio in the Bay Area in 2024. He has personally funded over $1M of the studio's experiments to date.

Studio

Acheul Inc.

Location

Palo Alto / Mountain View, CA

Founder

Taiga Sawamura

Contact

hello@acheul.com