nikJ13 - Overview

Hi there, I'm Niket Jain 👋

MS Computational Data Science @ CMU (GPA 4.03) | ML Intern @ Honeywell | NeurIPS 2025 Datasets Track | Building scalable LLM inference engines & agentic systems

🚀 Featured Experience

Role Key Impact Tech Stack
Honeywell ML Intern (Jun-Aug 2025) 2x faster sentence-transformer inference via ONNX + Kubernetes HPA; 35% better RAG w/ multimodal agent; Hackathon winner: 41% maintenance cost reduction ONNX, vLLM, Kubernetes, MCP
CMU Research Assistant (Prof. Carolyn Rose, Jan-May 2025) 40x speedup in image-guided code editing w/ QwenVL-32B agentic workflow; 150-instance dataset + SOTA baselines (0.7686 similarity) vLLM, QwenCoder-32B, GPT-4o
UBS Software Engineer (Jul 2022-Jul 2024) 87.5% faster doc processing w/ GPT-3.5 OCR/ETL; 60% data latency reduction via Kafka pipelines OpenAI APIs, Kafka, Java

🛠️ Core Skills

Languages & Databases

Python  •  Java  •  SQL  •  C++  •  Scala  •  R
MySQL  •  PostgreSQL  •  MongoDB  •  Redis

ML Frameworks & Infrastructure

PyTorch  •  Hugging Face  •  vLLM  •  TensorFlow  •  ONNX
Kubernetes  •  Docker  •  AWS  •  Ray  •  LangChain

DevOps & Tools

Git  •  SLURM  •  Helm  •  Terraform  •  FastAPI  •  Kafka
MLflow  •  Databricks  •  Modal  •  CUDA

🔥 Featured Projects

Inference Engine for Heterogeneous LLM Workloads

CMU, Dec 2025 | PyTorch, vLLM, Kubernetes

300 concurrent requests on 2x A100s with dynamic batching from first principles. Task routing: Self-Consistency (MMLU), Tool-use (Graph), Greedy (InfoBench).

Neural Network Backend Accelerator

CMU, Dec 2025 | C++, CUDA, Python

PyTorch-like framework with custom CPU/GPU backends for sparse ops. Optimized efficient GNN training and inference.

Data Attribution Benchmark for LLMs

NeurIPS 2025 Datasets Track | Python, Hugging Face, Ray

Benchmarked 8 data attribution methods across Pythia-1B to Llama-3.1-8B. Hugging Face leaderboard with 70% evaluation burden reduction.

Agentic Inference Systems

Jupyter Notebook | vLLM, LangChain, FastAPI

Scalable multi-agent orchestration framework for complex workflows and distributed inference.

📚 Education & Publications

Carnegie Mellon UniversityMS Computational Data Science (Aug 2024 - Dec 2025)

  • GPA: 4.0/4.0
  • TA: Mathematical Foundations of ML, Interactive Data Science
  • Coursework: LLM Methods, Deep Learning Systems, LM Inference, Cloud Computing

NeurIPS 2025Data Attribution Benchmark for LLMs (Datasets & Benchmarks Track) [Paper Link]

🌐 Let's Connect

LinkedIn Google Scholar Email


🔍 Open to ML Engineer roles in inference, distributed systems, & LLM infrastructure | Always excited to discuss deep learning systems and scalable ML solutions!

Last updated: December 2025