Currently work broadly in {privacy-preserving ML, sequential decision making in adverserial settings, dapp dev, cryptoeconomic simulators, credibility learning}, and have previously spent time in large-scale machine learning research + system design (cross-lingual NLP, active learning, async time series, knowledge graphs), latency-constrained modeling in the HFT space (online learning, real-time system engineering), and research across various subareas of machine learning in academia (mostly representation learning and human-in-the-loop learning).

Tech Stack: RabbitMQ, ZeroMQ, Celery, Redis, Hadoop, Spark, Airflow, Dash, InfluxDB, Postgres (+Timescale), Arctic, Datadog, PM2, Docker, Git, AWS, GCP, Linux (Ubuntu, CentOS, Fedora)

Research Areas of Interest: Privacy-preserving ML, Mechanism Design, Online Learning, Adaptive Control, Active Learning, Bandits, Differential Games

Languages: Python, C++, R, Bash/Fish, Javascript, Go, OCaml


I try to stay on the frontier. I aim for context-independence: my actions to be largely invariant under perturbations of my current environment (read: people + incentive structure). I constantly evaluate if I am doing something that is directionally in line with my long-dated goals + maximizing my knowledge-gradient, rather than focusing on what company, organization, or ecosystem I am in at a given time.

I am currently in areas where I focus on the symmetry, invariance, and reconciling why different solutions lead to the same outcome rather than finding one such path from question to answer, and moving on. I try to construct micro environments within problem spaces to extract feedback in a close-looped manner. I have always been fascinated by our —collectively— perception of reality rather than the ground truth itself, and how that manifests itself in our decision making in uncertain evironments.

I have no problem with picking up and starting over. Growing up, I've lived in twenty different states, and so far, have worked across a few different industries, although I'd argue I've been tackling largely the same set of problems. The changing environment has become engrained into who I am, and I welcome diving deep into something I have a high-variance prior over.

If I can help in anyway, or if you have a fairly low-variance prior that I should be doing something else, feel free to reach out.

Machine Learning

Worked across a variety of subareas of machine learning in academia + industry.

  • Engineered multi-tiered active learning system for entity resolution in a cross-lingual NLP (10+ languages) setting, built various text/speech based structures
  • Architected time series feature engineering and label engineering platform, representation learning/deep learning of multivariate time series, engineered a fast, sublinear, pattern search for petabytes of data
  • Developed novel community detection learning model on weighted bipartite graphs by constructing a cascade of learning tasks each leveraging structural properties of the graph
  • Co-designed, developed, and productionized recommendation system for ArXiv using a coaccess based learning approach, Joachims Lab
  • Researched jointly embedding heterogeneous signals in low-dimensional metric spaces via representation learning to trace vector paths through the latent space, Joachims Lab
  • Implemented Supersparse Linear Integer Model (SLIM), a mixed-integer optimization method [NP Hard] focusing on dimension reduction to produce interpretable scoring systems, used to analyze recidivism, MIT CSAIL
  • Designed, researched, developed (C++), and deployed multi-resolution online supervised learners in a low-latency environment in high frequency equities trading. Built a nonlinear discrete stochastic control framework, specifically focused on neuro-dynamic progamming (adaptive control) with state dynamics fueled by a lazy, nonparametric learner
  • Built novel online machine learners for multi-horizon ensembles, cascades, arbitrators, and label-partitioned DAG stackers with underlying structured output learners on high frequency, asynchronous time series + custom feature/label engineering


Too much to write here. Reach out. Work across the full-stack of the crypto space as it matures: governance, research, dapp dev, investing, defi protocol dev, trading.


Feel free to email me at akil.potti [at] gmail

If it's important, come find me in person