Burak Aksar

this is Burak.

CTO & co-founder. PhD in ML systems. I build production AI, end-to-end, not demos.

Work

  1. Spiky.ai

    CTO & Co-Founder

    2024-present

    Sales runs on data, but the feedback is always after the fact. We’re trying to give every rep a personal Jarvis: it listens, finds the patterns behind a win, and puts them to work live.

    Right now I’m deep into agentic graph architectures and self-improving loops: AI that gets better without manual intervention. Been studying Karpathy’s lecture on agents, loops, graphs, self-improving systems.

  2. IBM AI Research

    Research Scientist Intern

    2021, 2022

    Before LLMs were everywhere, I worked on making chatbot decisions explainable. Built a parser for multi-intent requests that matched deep learning accuracy at 48x the speed. Two patents came out of it.

  3. Sandia National Labs

    ML Engineer Intern

    2019, 2020

    Supercomputers slow down without warning. I built systems that catch the drift early, with very few examples to learn from.

  4. Boston University

    PhD, Computer Engineering · PeacLab

    2018-2024

    Big systems generate more data than people can read. I built ML that explains what’s wrong, not just that something is. Deployed it on a production supercomputer at Sandia. Fulbright Grantee, R&D 100 finalist.

Papers

2024 TPDS Runtime anomaly diagnosis in production HPC, with active learning
2023 arXiv TESS: multi-intent parser for conversational multi-agent systems
2023 SC Prodigy: unsupervised anomaly detection in production HPC code
2023 HPDC Practical ML frameworks for performance diagnostics
2022 BPM From natural language to workflows in RPA
2022 Cluster ALBADross: active learning for anomaly diagnosis code
2021 HPEC Network-data-driven allocation for HPC
2021 Euro-Par E2EWatch: end-to-end anomaly diagnosis code
2021 ISC Proctor: semi-supervised anomaly diagnosis code
2021 ICAPAI Counterfactual explanations for multivariate time series code
2020 SC Explainable ML frameworks for managing HPC systems
2019 ICPP HPAS: anomaly suite for reproducing performance variations code

Also

In my free time I’m trying to perfect espresso, and learning to produce electronic music, drums mostly. Always down for any form of volleyball: if your indoor league needs a sub, or you need an extra for grass, DMs are open. Feel free to send cocktail and restaurant recommendations, including what to try.

Hi