Devan Mallory

Location: [remote (US-based)] Type: Full-time or Part Time · Contingent on SBIR Phase I award Compensation: $150k-$225k year + equity About us We are an early-stage deep-tech company developing Cyquential AI, a neuro-symbolic architecture that induces grounded hierarchical symbolic structure directly from raw perceptual data — rather than relying on a hand-engineered knowledge base or bolting a symbolic layer onto a neural network. It uses a combination of Bayesian methods and Apriori algorithms to operate, working in a neurosymbolic-adjacent manner. The result is a cognitive architecture that learns from a handful of examples instead of millions, adds new tasks without forgetting old ones, and is fully interpretable end-to-end: every decision the model makes can be traced and read by a human. Our early work has demonstrated few-shot visual classification and continual learning with no deep learning, no pretraining, and no external data. The research was awarded first place at the Ohio State University Undergraduate Cognitive Science Symposium, with novelty cited by the judges as a primary reason for selection. We are now scaling the architecture toward interactive reinforcement learning and compositional reasoning, supported by an anticipated NSF SBIR Phase I award, as well as several other SBIR and Venture Capital Applications. The role You will be the senior research hire and work directly with the PI on the core architecture. This is a genuine research position — the open problems are real, unsolved, and yours to shape. We are open on seniority. We are hiring one-two people and will calibrate title, scope, and compensation to the candidate: a Research Scientist who can lead formalization and publication, or a Research Engineer who can build fast and run rigorous experiments. Strong candidates in either direction are encouraged to apply. What you'll work on Scaling the architecture to interactive RL environments (Arcade Learning Environment / Gymnasium) and to more challenging visual benchmarks including Mini-ImageNet and CLEVR-style visual question answering Designing and running rigorous evaluations: few-shot and continual-learning protocols, baselines, ablations, and sample-efficiency comparisons against established methods Strengthening the theoretical foundations of the system — the probabilistic semantics of its weight updates, the complexity and pruning behavior of its symbolic search, and its relationship to pattern mining and inductive logic programming Improving performance and scalability of the implementation, including a C++ backend for the search-heavy core Co-authoring publications for venues such as AAAI, NeurIPS, ICML, and IJCAI, and contributing to future SBIR Phase II and follow-on proposals What we're looking for Required MS or PhD in computer science, cognitive science, computational neuroscience, or a related field — or equivalent demonstrated research ability without the credential Strong Python; comfortable owning a nontrivial research codebase end to end Hands-on experience with at least one of: reinforcement learning, few-shot or continual learning, neuro-symbolic methods, inductive logic programming, probabilistic graphical models (Bayesian), or object-centric representation learning Demonstrated ability to design an experiment that could falsify your own hypothesis, and to report the result either way Comfort with ambiguity and with early-stage work where the roadmap changes as results arrive Nice to have Publications at ML or cognitive science venues (AAAI or symbolic conferences preferred) Experience with ALE/Gymnasium and standard RL evaluation protocols (Atari 100k, sticky actions, human-normalized scoring) Background in symbolic AI, cognitive architectures (SOAR, ACT-R, NARS), or Hebbian/spike-timing-dependent learning (STDP) C++ and performance optimization of search-heavy code Prior work on SBIR/STTR or other federally funded research You do not need experience training large deep networks. This is a different kind of system, and curiosity about why the dominant paradigm might not be the only one matters more here than fluency in it. Why this role Real scientific ownership. You will not be tuning someone else's pipeline. You will be shaping a novel architecture at the stage where the foundational decisions are still open. Publication is part of the job, not something you do on weekends. Ground-floor equity in a company built on research you helped produce. Advisory network. We are supported by mentors and advisors from the Shawnee State Level Up Accelerator programs, several of whom lead successful technology companies, and who remain actively engaged with the company. Important note on timing This position is contingent on an SBIR Phase I award (application submitted for NSF & DOW). We are being upfront about that rather than burying it: the role begins only if the award is made. We will keep every candidate informed of the outcome either way, and we are glad to talk before that decision is final — early conversations help us build a stronger proposal, and a named, committed collaborator strengthens the application itself. If you would rather engage before an award is confirmed, we are also open to a paid advisory or consulting arrangement at a small percentage of effort, which can convert to the full-time role.