Join the AI Research & Solutions Group
We're looking for an engineer who builds the models, not just uses them. Someone who has trained neural networks from scratch, constructed datasets that actually work, and implemented papers before the libraries caught up. You understand AI deeply because you've debugged gradient flows at 2am and know why your loss function is misbehaving. As a Lead AI Engineer in the AI Research & Solutions group, you'll work hands-on at least 3 days a week in the office in Menlo Park, leading the development of ML systems that power enterprise AI agents—systems that reason, act, and recover across long-horizon workflows. This is not a research scientist role. This is for the engineer who wrote the code that made the paper possible—the one who turned theory into working systems that ship.
What You'll Do
Build & Train Models
- Own the full pipeline: architecture design → dataset engineering → distributed training → production deployment
- Implement state-of-the-art techniques from papers—often before official implementations exist—translating math into working PyTorch code
- Build experimentation infrastructure that enables rapid iteration on architectures, training regimes, and evaluation methods
Build Agent Systems
- Create orchestration loops that enable AI to reason across multi-file changes, invoke tools (compilers, test runners, debuggers), and recover from failures
- Implement RL for code generation: execution-based rewards (RLVR), process supervision, reward models, and execution semantics alignment
Ship Production AI
- Optimize for inference: quantization, distillation, pruning, and serving infrastructure—understanding quality/latency/cost tradeoffs
- Design evaluation pipelines that catch regressions and measure what actually matters; build guardrails and monitoring for production
- Define integration architecture: APIs, batching, caching, failure handling
Lead Technically
- Mentor engineers on ML fundamentals, debugging techniques, and the craft of building reliable systems
- Stay current with the field—read papers, run experiments, share findings, separate signal from hype
What You Bring
Required (5+ years building AI that shipped)
- Trained real models: Architecture design through deployment—not fine-tuning tutorials, actual model development
- Deep ML fundamentals: Backprop, attention, loss functions, optimization—at a level where you can debug them
- Production engineering: Expert Python, PyTorch fluency, distributed training, GPU programming, ML infrastructure at scale
- Dataset intuition: You've built datasets and can spot data issues from model behavior—you know data quality beats model size
- Agent architecture: Built systems that plan, act, and adapt—agentic loops, tool-calling, multi-step reasoning pipelines
Strong Signals
- Implemented papers from scratch before official code existed
- Built or improved custom training infrastructure and experimentation pipelines
- Worked on LLMs, diffusion models, or frontier architectures at a low level
- Built RL training pipelines with non-standard rewards (execution-based, verification, process rewards)
- Implemented code retrieval systems: RAG for code, AST-aware search, repository-level indexing
- Worked on SWE-Bench or similar and understand what makes coding agents succeed or fail
- Open-source ML contributions or published implementations
Paths We Value
- Research backgrounds—PhDs, open-source contributors, or anyone who built models behind research
- Non-traditional paths from physics, math, or quantitative fields; show us models you've trained, not credentials
Who Thrives Here
- Builders, not talkers: You'd rather spend a week implementing a paper than writing a proposal about it
- Curious and rigorous: You ask "why does this work?" and don't stop until you understand
- Fast learners: New techniques come constantly; you pick them up quickly because you understand fundamentals deeply
- Pragmatic engineers: You know when to use existing solutions and when to build custom—optimize for shipping, not elegance
Why This Role?
Production AI systems that matter—not demos, not POCs. A team building next-generation AI agents for enterprise applications on OutSystems, where technical depth is valued over impressive-sounding credentials. If you've been frustrated that your team didn't understand why data quality mattered more than model size, this is your team. The Longer Story: OutSystems enables enterprise teams to build AI-powered applications and agents that reduce manual work, streamline internal operations, and accelerate impact. A proven low-code foundation combined with agentic AI and AI app generation capabilities empowers teams to move up to 10x faster with the assurance of security, scalability, an