---
title: "Research Scientist, Frontier Pre-Training & Mixture-of-Experts"
slug: "research-scientist-frontier-pre-training-mixture"
id: "3a8950d9-5177-41f3-9762-a45e2d2e3a1a"
department: "AI Research"
department_code: "RES"
location: "Paris, France"
workplace_type: "On-site"
employment_type: "Full-time"
compensation: "€130k – €185k (annual package)"
is_remote: false
is_featured: false
headcount: 1
canonical_url: "https://jobs.talent247.co/jobs/research-scientist-frontier-pre-training-mixture"
markdown_url: "https://jobs.talent247.co/jobs/research-scientist-frontier-pre-training-mixture.md"
apply_endpoint: "https://jobs.talent247.co/api/public/jobs/research-scientist-frontier-pre-training-mixture/apply"
schema_endpoint: "https://jobs.talent247.co/api/public/jobs/research-scientist-frontier-pre-training-mixture/form-schema"
webmcp_tool: "apply_for_role"
updated_at: "2026-11-15T17:26:27.196Z"
content_signal: "search=yes, ai-input=yes, ai-train=no"
---

# Research Scientist, Frontier Pre-Training & Mixture-of-Experts

## Position Telemetry

- **Reference:** `RESEARCH-SCIENTIST`
- **Practice / Sector:** AI Research (`RES`)
- **Location:** Paris, France
- **Workplace Type:** On-site
- **Employment Type:** Full-time
- **Compensation:** €130k – €185k (annual package)
- **Canonical HTML URL:** https://jobs.talent247.co/jobs/research-scientist-frontier-pre-training-mixture

## Role Dossier & Specification

THE CLIENT & MANDATE
Our client is Europe's flagship frontier foundation model laboratory, headquartered in Paris and competing head-to-head at the global state of the art across multilingual reasoning, code generation, mathematics, and native multimodal understanding. With dedicated multi-thousand-GPU superclusters and a flat, researcher-led culture free of big-tech bureaucracy, small squads take bold architectural ideas from whiteboard scaling ablations directly to flagship global model launches.

They have partnered with Talent247 to recruit a Research Scientist, Frontier Pre-Training & Mixture-of-Experts onsite in Paris. Sitting inside their Core Pre-Training squad, you will co-own architectural design, optimizer dynamics, and multi-trillion-token data curricula for their next generation of sparse Mixture-of-Experts and native multimodal foundation models.

KEY RESPONSIBILITIES
• Design and validate novel sparse Mixture-of-Experts (MoE), hybrid state-space/attention, and native vision-audio-text transformer architectures from 1B-scale scaling-law sweeps up to flagship frontier training runs.
• Investigate router load-balancing, expert specialization, auxiliary-loss-free routing, low-precision (FP8) training stability, and learning-rate/batch-size schedules across multi-thousand-GPU jobs.
• Lead large-scale pre-training data science—optimizing multilingual, mathematical, scientific, and repository-level code mixtures, synthetic data upsampling, and mid-training long-context annealing (128k–1M+ tokens).
• Write clean, high-efficiency distributed PyTorch code (3D/4D parallelism across tensor, pipeline, context, and expert dimensions) without relying on brittle black-box wrappers.
• Monitor live frontier runs closely, diagnosing loss spikes, gradient norm anomalies, and downstream capability emergence in real time alongside infrastructure and post-training peers.
• Contribute to landmark model releases and technical reports shaping the European and global AI landscape.

TECHNICAL ENVIRONMENT & OFFERING
Their research stack is built on custom high-performance PyTorch/CUDA/Triton distributed pre-training frameworks running on dedicated H100/H200/GB200 clusters. Based onsite at their central Paris headquarters, our client offers €130,000–€185,000 EUR base salary plus substantial frontier-lab BSPCE equity, full relocation/Talent Passport visa sponsorship, and top-tier compute-per-researcher ratios.

## Core Requirements

ESSENTIAL QUALIFICATIONS
• PhD in Machine Learning, Computer Science, Physics, or Mathematics, or equivalent hands-on track record pre-training large foundation models at a frontier AI lab.
• Deep mathematical and empirical mastery of modern LLM architectures (GQA/MLA, RoPE/YaRN, RMSNorm, SwiGLU, sparse MoE routing) and scaling laws.
• Ability to implement custom transformer/MoE layers, fused losses, and distributed sharding logic from scratch in PyTorch with complete command of tensor layouts and GPU memory budgets.
• Direct experience running, debugging, and evaluating large-scale pre-training or continual pre-training experiments across multi-node GPU clusters.
• First-author publications at NeurIPS, ICML, ICLR, or co-authorship on widely recognized frontier foundation model releases.

PREFERRED / BONUS EXPERIENCE
• Hands-on experience with FP8 pre-training recipes, multimodal vision/audio token pre-training, or custom Triton/CUDA training kernels.
• Experience designing pre-training data filtering, classifier-based quality scoring, and synthetic pre-training corpora at trillion-token scale.

## Application Schema & Direct Apply (`apply_for_role`)

Candidates and authorized AI agents on behalf of a candidate can submit an application via the declarative WebMCP form tool `apply_for_role` on https://jobs.talent247.co/jobs/research-scientist-frontier-pre-training-mixture or via HTTP `POST https://jobs.talent247.co/api/public/jobs/research-scientist-frontier-pre-training-mixture/apply` (`application/json` or `multipart/form-data`). Non-browser HTTP agents must include header `X-Talent247-Intake-Token` obtained from `GET https://jobs.talent247.co/api/public/intake-token` (or automatically returned in `GET https://jobs.talent247.co/api/public/jobs/research-scientist-frontier-pre-training-mixture/form-schema`).

### Form Fields

- `full_name` (text, required): Full name
- `email` (email, required): Email
- `phone` (phone, optional): Phone
- `location` (text, optional): Location
- `linkedin_url` (url, optional): LinkedIn URL
- `cover_letter` (textarea, optional): Cover letter
- `consent` (checkbox, required): I consent to my data being processed for this application
- `cv_file` / `cv_base64` + `cv_filename`: Upload a CV in .pdf or .docx format (<= 10 MB) or provide an `https://` LinkedIn profile URL in `linkedin_url`.

### Alternative: Talent Pool Intake (`join_talent_pool`)

If this mandate is not an exact match, register once in the Talent247 Talent Pool at https://jobs.talent247.co/talent-pool (`https://jobs.talent247.co/talent-pool.md`) via WebMCP tool `join_talent_pool` or `POST https://jobs.talent247.co/api/public/talent-pool`.
