Dolores Research Forward-Deployed AI Lab
San Francisco ↔ China

Most AI never reaches the work.
We embed until it does.

A founder-led, forward-deployed AI lab. We embed a senior team of builders and researchers into your environment and ship frontier AI into production — in weeks, not quarters.

Inside the work Watch · The interview
The Team’s Lineage

The model stopped being the bottleneck. The last mile is.

MIT's NANDA initiative found 95% of enterprise AI pilots produce no measurable impact — not because the models are weak, but because making them work inside one company's real data, workflows, and constraints is an engineering problem you can only solve from the inside.

We deploy upstream of your roadmap, not downstream of a contract. We build the system, operate it, and hand you the keys.

No decks. No rented hands. The thing that actually runs.

01

Last-Mile Diagnostic

2–3 Weeks

A senior founder embeds, maps your highest-value workflow, and returns a working prototype against your real data. Not a slide. A v0.

02

Embedded Build Sprint

60–90 Days

We deploy a senior team into your environment to design, build, and ship one production system end to end: evals, guardrails, monitoring, runbook.

03

Build · Operate · Transfer

90–180 Days

We run it in production, then transfer the code, the tooling, and the capability to your team. You end up self-sufficient.

04

Frontier Retainer

Ongoing

A standing line into frontier-model fluency: migrations, new capabilities, eval maintenance. Systems that improve as the models improve.

The category is no longer a bet.

In May 2026, OpenAI launched a $4B+ deployment company, Anthropic stood up a $1.5B enterprise services firm with Blackstone and Goldman Sachs, and ServiceNow, Accenture, and EPAM each launched forward-deployed engineering programs.

The frontier labs decided how enterprise AI gets deployed. We bring that model to founders who can't wait for a 10,000-person firm to show up.

A bridge most teams can't build.

We're fluent in the frontier models you can't easily reach, and the Western users you're trying to win. For ambitious teams going global, we bring three things a local vendor can't: compliant frontier-model access, SF-grade product taste, and a real read on US and EU markets. We architect so your sensitive data stays onshore, and contract through Singapore.

Lucid Labs cross-chain infrastructure platform Case 01
Lucid Labs — the infrastructure behind $150M+ in volume Lucid Labs · Financial Infrastructure · 2024
SP4C3 — decentralized AI compute-grade storage network Case 02
SP4C3 — a decentralized AI compute & storage network SP4C3 · AI Cloud + Distributed Storage · 2021
Case 03 Sentient Arena · Winner
71.5%
OfficeQA benchmark · built on MiniMax
+3.7%
Accuracy vs Claude Opus 4.5
1/500th
The cost vs Claude Opus 4.5
A frontier result at 1/500th the cost Sentient Arena · OfficeQA Benchmark · 2026

Consultants hand you a recommendation and leave. We hand you a system and the keys.

Staff-aug rents you hands. We own the outcome.

A big firm sends juniors and a statement of work. We send the senior people who sold you.

A team of founders, builders, and researchers who embed and build alongside you.

The founders presenting at MONAD × OpenBuild, Mushanghai
In the field · MONAD × OpenBuild · Mushanghai
The founders with deployed hardware at the Tuya developer booth
Shipping on-site · Tuya Developer · Mushanghai

Armin Tomas

Co-founder

A career CTO with a decade of shipping production systems for international clients.

Leon Liu

Co-founder

Ex-founder; advised startups and the Fortune 500, Nike among them. Now doing AI research in San Francisco.