Domain AI Researcher

Wipro•San Francisco, CA

2d•Remote

About The Position

Static datasets are no longer sufficient to evaluate frontier models; they must interact, plan, and execute within dynamic environments. We are seeking a highly technical Domain AI Researcher to build the ground truth for agentic workflows. You will design secure execution sandboxes, generate complex synthetic data, and identify the exact thresholds where models fail in multi-step reasoning tasks.

Requirements

Degree in Computer Science, Systems Engineering, or a related field with a heavy emphasis on machine learning and software architecture.
Advanced programming proficiency (Python, C++, or Rust) coupled with deep knowledge of Docker, CI/CD pipelines, and code-parsing tools (AST, Tree-sitter, LSP).
Hands-on experience with LLM orchestration frameworks (e.g., LangChain, AutoGen) and observability/tracing tools.
A strong investigative mindset tailored to finding edge cases—such as context window retrieval failures or infinite agent loops—and engineering the datasets required to resolve them.

Responsibilities

Build Execution Sandboxes: Architect secure, containerized environments (Docker, gVisor) to safely evaluate LLM-generated code, shell commands, and agentic actions across multi-file repositories.
Drive Data Synthesis: Build pipelines for rejection sampling, self-instruct, and synthetic data generation to create high-signal SFT and RLHF data specifically tailored to coding and system-level tasks.
Create the Ground Truth: Move beyond standard unit tests to curate complex edge-case scenarios, design AST-level validation checks, and measure functional accuracy on multi-step frameworks.
Collaborate & Publish: Partner with the Lead Researcher to transform execution sandboxes into publishable, open-source benchmarks (analogous to specialized SWE-bench frameworks) and co-author resulting research papers.