Hovhannes Grigoryan

About

3 min read

About

I am a Senior Applied Scientist at Amazon in New York City, working on causal inference, mechanism design, and agentic AI for large-scale decision systems. Before Amazon I spent over a decade at the intersection of theoretical physics, quantitative finance, and applied machine learning, a path that now anchors my research agenda:

Causal identification and mechanism design for agent-driven decision systems.

Path

I did my PhD in physics at Louisiana State University and Thomas Jefferson National Accelerator Facility (2008), with postdoctoral work at Argonne National Laboratory and the University of Chicago (Director’s Postdoctoral Fellow, 2008–2010), Ohio State University (2010–2011), and New York University (2011–2014). My theoretical physics publications, 16 peer-reviewed articles, 721+ citations, h-index 10, covered nuclear and hadronic physics. Representative work is indexed on Google Scholar.

After transitioning out of academic physics, I spent a year as a quantitative analyst at Citigroup Global Markets in New York, building pricing models for mortgage-backed securities and prepayment behavior. I then went through Insight Data Science and spent five years as adjunct faculty at CUNY, Manhattan College, and Georgetown, teaching calculus-based physics and advanced topics, running deep-learning research projects on Ising-model-class systems, and building financial forecasting tools on the side (winning the Quantopian Trading Contest four consecutive months in 2019).

I joined Amazon in August 2024 as a Senior Applied Scientist (L6) on the Creators Science team, where I now work on causal inference and LLM-agent systems for large-scale decision problems.

Research interests

  • Causal identification on panel data. Documenting where standard methods (two-way fixed effects, vanilla causal forests) fail, and building methods that don’t.
  • Mechanism design for learning agents. Proper scoring rules, incentive-compatible elicitation, auction-theoretic foundations for RLHF-style alignment.
  • Agent reliability engineering. Production patterns for LLM agents: tool-use, structured outputs, conformal wrapping for calibrated deferral.
  • Quantitative finance and time-series forecasting. Long-standing interest in mean-reversion, cointegration, regime-switching, and the honest limits of predictability in financial markets.

Skills

Programming. Python (primary), C++, R, Matlab, Mathematica, SQL, JavaScript, LaTeX.

ML libraries. PyTorch, TensorFlow, scikit-learn, AutoGluon, statsmodels, CausalML, EconML, doubleml.

Data & cloud. AWS (EC2, ECS, EMR, SageMaker, Bedrock, Cradle), Docker, Git, Spark/PySpark, Hadoop, DataGrip.

Domains. Causal inference (DML, causal forests, DiD, QTE, synthetic control), time-series forecasting (TFT, N-BEATS, MQF, classical econometrics), reinforcement learning (model-based and model-free), LLM agents (Bedrock, LangChain, MCP), generative models (GANs, VAEs, diffusion, flows).

Licenses. FINRA Series 7 and 63 (historical; expired).

Languages. English, Russian, Armenian (fluent).

Selected achievements

  • 16 peer-reviewed physics publications, 721+ citations, h-index 10.
  • US Patent Application 2006-0044564 (published as US 2007/0149866).
  • 9 invited talks at national-scale conferences; co-organizer of 2 scientific conferences.
  • Quantopian Trading Contest winner, 4 consecutive months (2019).
  • Director’s Postdoctoral Fellowship, Argonne National Laboratory (2008–2010).

Contact

Open to conversations with researchers at MSR, Google Research, academic departments, and applied-science groups working at the intersection of causal methodology, mechanism design, and learning agents.

Graph View

Backlinks