cover_letter / art_xdvZ2_gBs8U

Dear Mistral AI Hiring Team, Mistral AI occupies a rare position in the AI landscape: building frontier models that are simultaneously high-performance, open, and genuinely deployable in enterprise production environments. That combination — research rigor paired with real-world utility — is what drew me to apply. My own trajectory, from hand-coding backpropagation through time in C++ at UC Berkeley in 2004, to publishing at NeurIPS, to shipping multi-agent orchestration frameworks and RL post-training workbenches today, reflects the same conviction that AI's value is realized only when it runs reliably in production. **Technical Foundation** My AI/ML work spans the full stack from research to deployment. At the research end, my NeurIPS 2014 paper on artificial neural networks for protein secondary structure prediction established early credibility in deep learning — work that originated from a hand-coded C++ neural network with custom BPTT, later rewritten in PyTorch scaling from 413 to 8 billion parameters across five architectures (feedforward, GRU, Transformer, ESM-2, multi-task), with MLflow experiment tracking, Optuna hyperparameter optimization, and FastAPI serving. More recently, I built an RL post-training workbench covering the full RLHF/DPO pipeline: a Reward Lab for designing and A/B testing reward functions (RLVR, learned, hybrid) across GSM8K, MATH, HumanEval, and UltraFeedback; a Playground for real TRL-powered GRPO/DPO training with live SSE metric streaming on Apple Silicon (MPS) and CUDA; and an Arena for head-to-head framework benchmarking across TRL, VeRL, OpenRLHF, and NeMo RL with GPU passthrough in Docker containers. I implemented 12 RL algorithms — PPO, GRPO, DAPO, REINFORCE, REINFORCE++, RLOO, DPO, SimPO, IPO, KTO, ORPO, and SPPO — with standardized throughput, memory, and convergence benchmarking across frameworks. On the evaluation side, I built aeval, a local-first model evaluation platform with five core eval types (factuality, reasoning, instruction-following, safety, code generation), adversarial safety testing with refusal detection, bootstrap confidence intervals, Welch's t-test, Cohen's d effect size, and CI/CD integration with automated safety gates — running on a FastAPI orchestrator, TimescaleDB, Redis job queue, and Next.js dashboard backed by Ollama. For agentic and RAG applications, I architected the Fintellect RAG retrieval pipeline with ChromaDB vector store, multi-provider LLM orchestration across Claude, GPT-4, and Gemini with fallback routing, structured output validation, and token budget optimization. I also built OpenClaw, a multi-agent orchestration framework with gateway protocol, subagent delegation, profile management, and session switching — enabling coordinated AI agent workflows across multiple industry verticals. **Why This Role** My career has consistently sat at the intersection of deep technical work and customer-facing deployment — as a Staff PM at Intuit scaling developer platforms to 675M+ engagements, as a founder shipping production AI applications, and as an adjunct faculty member translating complex ML concepts for students. The Forward Deployed ML Engineer role at Mistral is precisely the convergence of those threads: working directly with enterprise customers from pre-sales through production, while contributing back to the models and open-source infrastructure that make those deployments possible. What excites me specifically about this role is the scope of the customer engagement — managing relationships across CEO/CTO, data scientists, and software engineers simultaneously, which maps directly to how I operated at Intuit and Splunk, where I regularly translated between executive strategy and engineering implementation. The opportunity to externalize Mistral's research into production settings, and to feed customer signal back into product and model improvements, is exactly the feedback loop I find most generative. **Selected Relevant Experience** - **RL Post-Training Workbench (2026):** Implemented 12 RL algorithms with cross-framework benchmarking (TRL, VeRL, OpenRLHF, NeMo RL), live SSE metric streaming, and GPU Docker passthrough — directly applicable to guiding customers through fine-tuning and post-training workflows on Mistral models. - **aeval Platform (2025–2026):** Built end-to-end model evaluation platform with adversarial safety testing, statistical rigor (bootstrap CI, Welch's t-test, Cohen's d), and CI/CD regression detection — relevant to the evaluation guidance responsibilities in this role. - **Fintellect AI RAG Pipeline:** Architected multi-provider LLM orchestration with ChromaDB, fallback routing, structured output validation, and token budget optimization — directly applicable to advanced RAG use cases Mistral customers will bring. - **OpenClaw Multi-Agent Framework:** Built production multi-agent orchestration with gateway protocol and subagent delegation, enabling coordinated AI workflows — applicable to agentic use cases referenced in the JD. - **Intuit — ICE Platform (675M+ engagements, FY23):** Delivered self-service developer platform reducing onboarding from 2–3 weeks to minutes, scaling throughput from 6K to 50K TPS via rSocket migration; extended Java and Python SDK Starter Kits with scaffolding, CI/CD integration, and testing frameworks — directly relevant to onboarding customers on Mistral APIs and production integrations. - **Splunk — Search Orchestration (2019–2021):** Owned Go microservices, PostgreSQL metadata services, and SPL/SPL2; delivered Scheduler Service end-to-end in ~4 months; achieved up to 10x query performance improvements for enterprise beta customers — demonstrates ability to move quickly on complex technical deliverables with direct customer impact. - **NeurIPS 2014 — Protein Structure Prediction:** Published research on neural networks for secondary structure prediction, with the 2026 PyTorch rewrite spanning 413 to 8B parameters — establishes the deep learning foundation the role requires. **Closing** Mistral's mission to democratize AI through open, high-performance models is one I find genuinely compelling — not as a slogan, but as a structural bet that the field benefits from capable models that enterprises can actually inspect, fine-tune, and deploy on their own terms. I want to help more organizations realize that bet in production. I would welcome the opportunity to discuss how my background in RL post-training, RAG and agentic architectures, developer platform scaling, and enterprise customer engagement maps to what your Applied AI Engineering team is building. Thank you for your consideration. Sincerely, **O. Felix Amoruwa** famoruwa@berkeley.edu | 909-731-9011 | felixamoruwa.info