Arrakis (Arrakis Technologies)

Verified

Track: stealthWebsite ↗𝕏 @arrakistech

potential 72 · scoop 88

Arrakis builds an AI agent platform for industrial operations with model-agnostic orchestration and knowledge graphs. Three founders from Palantir and Accel secured backing from Accel.

Scores run 0 to 100. Potential: how likely the firm is to become a real winner, judged across the six dimension scores and held consistent with them, whether or not anyone has heard of it. Scoop: how unlikely a well-read tech investor is to have heard of it yet. High means it has not reached the outlets that reader follows, whatever the firm's quality or age. How our top potential scores did.

ShareXLinkedIn
Haroun Beltaïfa left his Palantir pod lead role after procurement deployments at Morson Group to co-found the company in early 2026.

Snapshot

What they build
Builds AI agents that execute real-world tasks for industrial enterprises.
Who buys
Industrial enterprises needing autonomous AI execution.
Comparables
X for Y analogs in agentic automation (e.g., similar to specialized industrial AI agents)
Sub-industry
AI agents for industrial/enterprise execution

Ex-role leading to industrial AI focus; credible operator transitioning from prior AI work.

How we found them

Surfaced in issue
Jun 29, 2026
Announced
Jun 25, 2026 (stealth reveal)
Founded
2026
HQ
United Kingdom
Discovery track
stealth

Originating post ↗

Dimension scores

  • Team: 78 out of 100
  • Product: 52 out of 100
  • Traction: 25 out of 100
  • Market: 72 out of 100
  • Moat: 38 out of 100
  • Capital: 68 out of 100

Analysis

Romain Fouilland led the Hub Santé health data platform at ANS for three years. He co-authored the arXiv paper 2210.10610 on quantum graph transformers. Rafael Quintanilla moved from Accel investing roles. Haroun Beltaïfa spent six years at Palantir in forward-deployed positions.

The platform runs model-agnostic orchestration over living knowledge graphs. It handles unstructured data from emails and PDFs. On-prem and SOC 2 options target defense and energy buyers. The team sells outcome-based contracts instead of seats.

Accel provided early backing. Founding engineer postings continue. No named customers or production metrics have surfaced. The current edge rests on founder networks rather than locked data or licenses.

What they do

Arrakis partners with industrial enterprises to design, build, and scale model-agnostic AI agents that execute real-world operational tasks (e.g., supplier risk hedging, vessel monitoring, spend analysis) via integrations, knowledge graphs, governance layers, and embedded vertical teams, with value-based pricing and on-prem/cloud deployment options for regulated environments.

Why now

Agentic capabilities (planning, tool use, multi-step execution) matured sufficiently post-2023/2024 foundation models to move beyond chat/pilot demos, yet industrial enterprises report <10% production scaling due to integration, unstructured data, governance, and execution bottlenecks, exactly the gap the platform targets. Concurrent regulatory and efficiency pressures in defense, energy, and manufacturing plus pricing accessibility of capable models enable the “weeks not quarters” delivery claim.

Founding story

Founders connected through overlapping French elite technical networks (Mines ParisTech/Corps des Mines alumni) and Palantir’s forward-deployed enterprise AI deployment culture. Romain exiting multi-year government/health data platform role at ANS; Rafael brought recent Accel investing experience; Haroun’s Palantir exit timing aligns with founding window (late 2025/early 2026 incorporation). Origin narrative via stealth reveal post and job descriptions emphasizing “AI-native OS for the physical economy.”

Competitive position

Obvious incumbents: Palantir (AIP platform; team has intimate knowledge via alumni) and broader industrial software players (Siemens, GE, SAP with AI overlays). Closest startup competitors: emerging agent platforms focused on enterprise execution or vertical AI. Positioned in the same market with a differentiated wedge, model-agnostic + embedded industrial research teams + physical-world focus + fast, governed deployment, versus Palantir’s broader data platform or horizontal agent tools.

Defensibility

Primary signals are team expertise (Palantir forward-deployed deployment know-how, vertical research) and network access rather than proprietary data lock-in or regulatory licenses at this stage. Knowledge graph + governance layers could create switching costs over time in mission-critical workflows. OSS/community ownership absent. Overall moat is early and talent/network-dependent; replicable core architecture (model-agnostic orchestration) lowers barriers for well-resourced competitors.

Network density

At least 2, 3 of the named early team (Haroun + references to “former Palantir … engineers”) came from Palantir’s forward-deployed/pod model; additional French Grandes Écoles density (Mines ParisTech/Corps des Mines for Romain and Haroun; Imperial for Rafael) and Accel network ties. Fraction: multiple early technical/leadership roles traceable to Palantir + elite French engineering corps.

Early team signals

Lead/backing signal from Accel (job postings explicitly “Accel-backed”). Board/advisors: none publicly named. Relationships: Palantir alumni network provides direct access to enterprise deployment playbooks and potential customer cohorts in complex ops (defense, procurement, industrial). DeliveryHero engineering experience referenced in hiring materials for operational scaling.

Traction

  • Accel backing
  • Founding-engineer job postings
  • Stealth reveal X post

Opportunities

  • Maturation of agentic tooling enabling reliable multi-system execution in high-stakes environments
  • Defense and industrial modernization budgets prioritizing AI for operational resilience and supply-chain visibility
  • Regulatory tailwinds around explainability, auditability, and governance in high-risk sectors (aligns with platform’s SOC/ISO/GDPR and traceability emphasis)
  • Accel network for enterprise intros and follow-on capital
  • Potential wedge expansion from health-data integration experience (Romain) into adjacent regulated verticals

Risks / red flags

  • Extremely early stage (stealth reveal, placeholder “0+” metrics on site, active founding hires) with no independently confirmed paying customers or usage data
  • Heavy embedded-services component risks low scalability vs. pure product
  • Defense/aerospace exposure carries potential ITAR/export-control or high-risk AI regulatory burdens (EU AI Act) not visibly detailed in public materials
  • Network homogeneity (Palantir + French elite engineering corps + Accel) may limit operational diversity under pressure
  • Unverified technical/performance claims on website
  • Funding details sparse beyond Accel association; incorporation signals (Companies House Dec 2025) indicate very recent formalization

Hidden bets

  • Model-agnostic architecture with explicit “bring your own model” + real-time optimization across providers
  • Heavy emphasis on unstructured data parsing (emails, PDFs, spreadsheets), living knowledge graphs, granular governance/auditability, and on-prem deployment for regulated environments (SOC 2, ISO 27001, GDPR)
  • Strategic bet on forward-deployed/embedded vertical teams + applied research (Palantir-style) rather than pure self-serve product
  • Value-based pricing tied to delivered outcomes
  • Non-obvious wedge: industrial/physical-world execution (aerospace/defense, energy, shipping) with “command center” orchestration

Key relationships

  • Palantir alumni network
  • Accel network

Founders

Romain Fouilland, Co-Founder (technical/AI focus)
Education: Corps des Mines / Mines ParisTech (Master in Business and Public Administration, 2019–2022), Master in Applied Maths, École Polytechnique (data science & AI major)
Key projects: arXiv 2210.10610 “Extending Graph Transformers with Quantum Computed Aggregation” (co-author), Surf Buddy (AI-powered surf training via video + LLM/CV)
Prior: ANS (Agence du Numérique en Santé) / Project Lead, Hub Santé health data platform / ~3 years (left recently)
GitHub
Rafael Quintanilla, Co-Founder & CEO
Education: Imperial College London
Prior: Accel / Vice President / Senior Associate / Associate (early-stage AI, fintech, consumer) / ~2022 until co-founding
𝕏 @RafaelFQL
Haroun Beltaïfa, Co-Founder
Education: MINES ParisTech (advanced Mathematics and Physics)
Key projects: Internal Palantir AIP deployments for enterprise operational efficiencies (e.g., procurement use cases with Morson Group)
Prior: Palantir Technologies / Forward Deployed Engineer / Pod Lead / Procurement Lead / ~6 years (left ~Jan 2026)

Citations