Parth K. Thaker

AI Research Engineer at Intuitive Surgical — building secure, privacy-preserving LLM systems: retrieval assistants over internal documents and code, guardrail and security analysis for LLM applications, and LLM + CI/CD tooling. Open to AI research engineer roles.

01
Bandit-based multi-agent search under noisy observations
P. Thaker, S. Di Cairano, A. P. Vinod
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.
IFAC World Congress 2023
2023
02
Maximizing and Satisficing in Multi-armed Bandits with Graph Information
P Thaker, M Malu, N Rao, G Dasarathy
A pure-exploration bandit method (GRUB and ζ-GRUB) that uses a similarity graph over arms to find the best arm, or one above a threshold, with far fewer samples than standard UCB.
NeurIPS 2022
2022
03
On the Sample Complexity and Optimization Landscape for Quadratic Feasibility Problems
P. K. Thaker, G. Dasarathy, A. Nedic
Proves identifiability and sample-complexity bounds for recovering a complex vector from quadratic measurements, and shows the nonconvex landscape lets gradient descent converge globally from arbitrary initialization.
ISIT 2020
2020

01
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.
LLMsRAGPython
02
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 bandits

›
RAG That Renders the Diagram: Outline-Aware Chunking and Image-Grounded Answers
How I built a hallucination-resistant RAG over a large technical-document corpus: chunk along the author's heading outline, embed with that outline as context, and treat images as first-class chunks with a deterministic verifier.
Jun 2026
›
The real cost of best-arm identification isn't the arms — it's the clusters
When arms come with a similarity graph and rewards vary smoothly over it, best-arm identification's binding cost drops from the number of arms to the number of clusters.
Jun 2026
›
The Branch Is the Variant: Zero-Copy, Privacy-Safe Test Data on a Versioned Lake
How a versioned data lake made the branch itself the test-data variant — zero-copy generation, blobstore-native metadata, and in-database hooks as the output channel.
Jun 2026