Open to AI research engineer roles

Parth K. Thaker

AI Research Engineer working on Nonconvex optimization

I apply optimization and bandit theory from my Ph.D. to building secure, applied LLM systems. Currently working on privacy-safe LLM workflows at Intuitive Surgical; previously research on graph-based bandits and nonconvex optimization at ASU, MERL, and IISc.

View research Projects Writing About / CV

Now

AI Research Engineer · Intuitive Surgical

Ph.D.

Electrical Engineering · ASU, 2024

Earlier

B.Tech / M.Tech · IIT Madras

Selected research

Optimization, bandits, and graph theory — several with open-source code.

All papers →

US Patent Application US20240427327A1

Controlling Search Agents to Perform Search with Noisy Observations and Probabilistic Guarantees

A control system that directs teams of search agents along fuel-constrained paths using a multi-level multi-armed bandit, classifying regions from noisy measurements with probabilistic finite-time guarantees.

Read

ICASSP 2024

Non-Stationary Bandits with Periodic Behavior via Ramanujan Periodicity Transforms

A multi-armed bandit method that uses the Ramanujan periodicity transform to detect periodic patterns in shifting rewards and exploit them to reduce regret.

Read

IFAC World Congress 2023

Bandit-based multi-agent search under noisy observations

A thresholding multi-armed bandit algorithm for coordinating multiple search agents under noisy sensor data, with finite bounds on search time, labeling time, and cost.

Read

Selected projects

Applied systems — from open-source research code to secure LLM tooling.

All projects →

Secure, privacy-preserving LLM workflows

Privacy-safe assistants and guardrails that let teams use LLMs over sensitive internal material without data leaving the trust boundary.

LLMsRAGPythonCloud

GRUB: graph-informed pure-exploration bandits

Open-source bandit algorithms that use a similarity graph over arms to find the best (or good-enough) option with far fewer samples than standard UCB.

PythonNumPyMulti-armed banditsGraph signal processing

Latest writing

All writing →

Projected Gradient Descent with Skipping

Exploring an interesting optimization technique that skips projection steps to potentially improve convergence in constrained convex optimization.