Lead Research Engineer

Kog is a European VC-funded startup and real-time AI frontier lab building the world’s fastest AI execution layer. As part of the 2030 French Tech cohort, we are on a mission to redefine the boundaries of artificial intelligence by enabling true real-time interaction at a scale never seen before.While the industry often settles for incremental software updates, we are engineering a radical, vertically integrated solution. Our approach is built on three deeply interconnected streams that form the core of our competitive advantage:GPU Engineering: We are building the Kog Inference Engine, a proprietary runtime purpose‑built for Dense and MoE LLMs. We develop our own kernels directly in Assembly on AMD Instinct accelerators, entirely bypassing standard libraries to extract the theoretical maximum throughput from the hardware.Model Architecture: We do not just run models; we reinvent them. Our researchers design model architectures specifically optimized for our engine, creating a hardware-software co‑design loop that enables massive performance leaps.We aim to achieve 100x faster token generation compared to current industry standards, targeting 10,000+ tokens per second to unlock truly instantaneous AI.While our GPU team extracts maximum throughput from the hardware, we know that raw compute is only half the equation. To reach this scale, we must fundamentally rethink how models are built. We are moving beyond standard architectures to create models that are natively designed for our execution engine.We are achieving this through Hardware-Software Co‑design, focusing on breakthroughs in:Architecture‑Hardware Alignment: Designing model layers and dimensions that map perfectly to our proprietary kernels and memory hierarchy.Next‑Gen Architectures: Moving beyond standard Transformers to explore and implement linear attention, SSMs (Mamba), and highly optimized Mixture-of-Experts (MoE).Extreme Sparsity & Quantization: Implementing aggressive quantization‑aware training and structured sparsity to minimize memory bandwidth usage without sacrificing intelligence.Training for Inference: Designing training pipelines that directly prioritize final inference velocity as a core objective.Our work is recognized by industry leaders, as evidenced by our recent performance benchmark , which was published on AMD’s official blog.Now, we need the architectural vision to multiply that leverage.The Culture:You will join a high‑intensity environment where velocity is a core virtue.We treat research as an iterative engineering process: we prioritize execution cycles over theoretical perfection, shipping over talking, and technical truth over corporate consensus.At Kog, you will not just be an engineer, you will be a pioneer building the foundational infrastructure for the next generation of collaborative, real‑time AI agents.What You’ll DoWe are seeking a Lead Research Engineer with strong managerial experience or senior expertise to act as a strategic partner to the CEO.You will bridge the gap between high‑level research vision and concrete execution, turning ambitious ideas into a production‑ready reality. Your role is a hybrid of high‑impact technical leadership and hands‑on engineering.You will be expected to:1/ Technical Strategy & Execution (The "Owner")Own the Roadmap: Take high‑level scientific directives and abstract concepts from the CEO to define concrete objectives and convert them into a structured, actionable research roadmap.Architect the 10x leap: Lead the engineering efforts to design and train our proprietary models, ensuring they are optimized for our inference engine and balanced for performance (quality vs. latency).Hardware‑Software Co‑design: Collaborate deeply with the GPU Engineering stream to influence model design. You will provide the feedback loop that allows us to structure models specifically to exploit our latest kernel optimizations and memory hierarchy breakthroughs.Accountability: You are fully accountable for the delivery. You ensure that we don't just have a plan, but that we execute it with precision and velocity.2/ Hands‑on Engineering (The "Expert")Strategic Contribution: You are capable of diving into the training codebase to unblock the team or tackle critical architectural challenges when necessary.Lead by example: You maintain a deep understanding of the stack (Training Infra, PyTorch/JAX, Distributed Systems) to conduct code reviews and guide technical decisions, ensuring quality without being the bottleneck.Deep‑dive optimization: Spearhead the resolution of complex training issues (convergence stability, distributed training efficiency, data pipeline bottlenecks) to ensure our experiments are rigorous and fast.System‑level creativity: Leverage your deep understanding of Deep Learning and Hardware to find architectural solutions that maximize our specific inference engine capabilities.3/ Team Leadership & Culture (The "Captain")Manage and Coach: Manage a team of researchers and engineers (~5 people). You act as a "Head Coach" with the mandate to compose your squad: you define the roles, mentor high‑performers, and make necessary adjustments to ensure the team meets our high standards.Structure & Buffer: You act as the interface between the CEO and the team. You clarify priorities, filter the noise, and absorb the pressure to allow your team to focus on execution while ensuring deadlines are met.Entrepreneurial Ownership: Act not just as an employee, but as a builder of the company. You instill a startup mindset, favoring rapid iteration and concrete results over pure academic exploration.Whom we'd like to work withWe are seeking a unique profile: a deeply technical researcher/engineer who enjoys the craft of building models, but also possesses the maturity to lead a team and develop a roadmap. You are a "force multiplier" and make everyone around you better.1/ Technical Expertise:Training Authority: You have a PhD or a top‑tier Engineering degree with deep experience in training Large Language Models (LLMs) or complex architectures. You understand training dynamics, convergence stability, and distributed systems.Architecture Intimacy: You understand exactly how modern architectures work under the hood (Transformers, MoE, SSMs/Mamba). You are not just using libraries; you understand the mathematical and hardware implications of every layer.Engineering Rigor: Unlike pure academic researchers, you write robust, scalable code (PyTorch/JAX). You bridge the gap between "research code" and "production‑ready infrastructure."2/ Leadership & Accountability:Transform Directions into Concrete Plans: You can take high‑level, abstract scientific directives from the CEO and turn them into a concrete, executed engineering plan. You act as the bridge that structures the team's daily focus.Flexible Experience Level:Option A: You are already a Research Manager / Tech Lead with experience managing a high‑performance team.Option B: You are a Senior/Staff Researcher at a top‑tier lab or tech company, looking to take the next step in your career and shoulder managerial responsibilities."Head Coach" Approach: You manage with a focus on sustainable performance. You know how to compose your team (recruiting, adjusting roles) and channel pressure to get results without burning people out. You prioritize shipping over endless exploration.3/ Mindset:Superstar without the Ego: You are confident in your skills but humble in your interactions. You are "brilliant but mature", you prioritize the team's success over personal recognition.Entrepreneurial Drive: You understand the startup pace. You favor rapid iteration cycles and "good enough" prototypes over theoretical perfection. You treat the company as if it were your own.Results over Papers: While you value scientific rigor, your primary metric of success is not the number of citations, but the performance of the model in our production engine.What we offer:Top‑Tier Compensation: We offer a highly competitive salary package (top of the market) tailored to match your expertise and leadership level.Real Ownership (BSPCE): You aren't just an employee; you are a partner. We offer significant equity to ensure you share in the startup's success.Unrivaled Technical Playground: Work on the bleeding edge of AI hardware. You will have access to the compute power you need (high‑end clusters) to perform your magic.A world‑class Environment: Join a high‑density talent team of 12 engineers (including 5 PhDs). We value peer‑to‑peer learning, high autonomy, and zero bureaucracy.Impact & Autonomy: As a Lead, you will have a direct seat at the table to shape our engineering culture and roadmap alongside the CEO.Remote‑First & Team Bonding: We operate as a remote‑first company, valuing autonomy and deep work. Our culture is punctuated by our monthly "Paris Weeks" one week per month, where the whole team gathers at our WeWork offices in the 13th district (near Station F), the heart of Paris' tech scene. These weeks are dedicated to strategic alignment, intense collaboration, and team bonding.Ready to build the 10,000+ tokens/sec stack? Apply directly to start the conversation #J-18808-Ljbffr