Looking for a job

PRANSHU JAIN

@pranshu32

Message

Data Science Fellow and AI/ML researcher building agentic LLM systems and causal inference solutions.

India

Message

What I'm looking for

I’m looking for an AI-focused role where I can build reliable LLM/agentic systems, apply causal inference to improve decisions, and work with engineers to ship measurable improvements in performance, safety, and efficiency.

I’m a Data Science Fellow focused on applying causal inference and AI safety principles to real-world decision making. I’ve built systems that improve reliability and efficiency—using refuter methods to cut runtime, and causal frameworks to strengthen treatment-effect analysis.

In my current Co-op Intern role at AMD, I architect LLM-based test automation for GPU validation, optimize ROCm ML frameworks for throughput, and develop agentic AI workflows that reduce hallucination rates through iterative plan review. My research and projects span NLP, explainable AI, LLM fine-tuning (quantization), reinforcement learning (DPO/PPO), and kernel-level ML optimizations.

Experience

Work history, roles, and key accomplishments

Current

Co-op Intern (LLM GPU QA)

Current

AMD

Jul 2025 - Present (11 months)

Architected an LLM-based test automation agent for GPU validation, reducing manual optimization effort by 20% and improving edge-case defect detection by 10%. Optimized ROCm ML frameworks (Flash-Attention, hipDNN) for ZEN/CDNA architectures, achieving 4% throughput gains with mathematical parity via scaled kernel validation, and reduced application hallucination rates by 65% using agentic AI workf

GPU Validation ROCm HipDNN Multi Model Deliberation

Research Intern (AI Safety)

IIT Madras

May 2025 - Jul 2025 (2 months)

Designed EduBench, an AI safety framework integrating technical rigor with global policy to improve context-aware governance across 9 educational risk domains. Improved framework robustness by 28% and safety evaluation accuracy by 20% using multi-model testing and personalized harm perspective integration.

Benchmark Design Multi Model Testing Domain Evaluation Harm Perspective Modeling

Data Science Fellow (Causal)

Fellowship.ai

Sep 2024 - Dec 2024 (3 months)

Applied causal inference (Causica/DoWhy) to analyze retail datasets and enable treatment effect analysis that improved decision quality by 25%. Strengthened model reliability using refuter methods, reducing runtime by 12% and improving overall application efficiency by 10%.

Causal Inference Causica DoWhy Treatment Effect Modeling Refuter Methods Reliability Decision Quality

Open Source Developer

CodeforGovTech

Jul 2023 - Sep 2023 (2 months)

Optimized search performance by engineering an Algolia-style NPM package for SunbirdED-Search, reducing filter configuration time by 30% for 1cr+ users on the DIKSHA portal via real-time rendering. Improved end-to-end search responsiveness to support large-scale national usage.

Algolia SunbirdED Search DIKSHA Portal Real Time Rendering JavaScript (Node.Js Performance Engineering