# Jagrut Nemade

> AI Engineer at Aerospike | Ex-AI & ML Engineer at UW–Madison | Ex-Deloitte | MS Data Science | Python, R, SQL

Location: United States, United States
Profile: https://flows.cv/jagrut

## Work Experience
### Software Engineer @ Aerospike
Jan 2025 – Present | Mountain View, California, United States

### AI Engineer Intern @ University of Wisconsin-Madison
Jan 2025 – Jan 2025
1) Built an Agentic RAG assistant (LangGraph, LangChain, ChromaDB) used by 500+ radiology staff to select CT Imaging protocols
2) Optimized inference with vLLM batching, Asyncio, reducing latency by 40% for secure deployment in the private hospital network
3) Deployed as a FastAPI service with Docker. Reduced protocol lookup time by 60% through direct answers with source references

### AI Engineer - (Graduate RA) @ University of Wisconsin-Madison
Jan 2024 – Jan 2025 | Madison, Wisconsin, United States
1) Collaborated with Flywheel engineers to extract, preprocess 80K+ unstructured radiology reports for downstream analysis
2) Built an Agentic RAG system (LangGraph + Langchain + LLaMA) for medical QA, evaluated on 1K+ reports using ROUGE metric
3) Dockerized and Deployed as Flywheel Gear reducing lookup time by 70% and forming the base for the client’s multimodal RAG

### Teaching Assistant @ University of Wisconsin-Madison
Jan 2024 – Jan 2024 | Madison, Wisconsin, United States

### Machine Learning Engineer - (Graduate RA) @ Wisconsin Institute for Discovery
Jan 2024 – Jan 2025 | Madison, Wisconsin, United States
1) Tackled the resource-intensive task of labeling datasets and fine-tuning LLMs by designing Uncertainty-based and K-Center Clustering data selection strategies to identify the most informative prompts
2) Fine-tuned LLaMA-2 on CUDA GPUs, achieving 53% win rate (AlpacaEval), reducing training data by 67% & GPU runtime cost by 51.4%
3) Demonstrated accuracy comparable to full dataset fine-tuning, showing 3x efficiency gains without loss in model performance
4) Validated results through statistical hypothesis testing and monitored training with Weights & Biases to ensure reliability

### Data Analyst | ETL Pipelines & Data Processing @ Deloitte
Jan 2022 – Jan 2023 | Hyderabad, Telangana, India
1) Built SQL pipelines in Oracle ERP (Costing, Procurement) & Tableau/BI dashboards to deliver real time insights for client
2) Optimized ETL pipeline runtime efficiency by 5x through query tuning and automation for faster, reliable data processing
3) Led cloud migration of client data with Oracle Integration Cloud. Streamlined procurement workflows & improved planning decision

### Undergraduate Researcher @ University of Malaya
Jan 2022 – Jan 2022
1) Designed a pipeline with three components: U-Net for segmentation, a CNN architecture (based on ImageNet) for feature extraction, and Random Forest classifier, utilizing hypothesis testing to reprocess ambiguous images

2) Conducted extensive experiments and statistical analysis to validate the proposed architecture, successfully testing it to segment people and their attire from images and classify the attire based on regional origin which can be utilized for photoshopping applications.

3)Evaluation: Conducted ablation studies to evaluate individual components and assessed overall performance using F1 score, achieving a 10% accuracy improvement in image classification through the feedback-driven segmentation loop

### Undergraduate Research Assistant @ Indian Institute of Information Technology Dharwad
Jan 2021 – Jan 2021
1) Classified sentences as Hope Speech, Non-Hope Speech, or Neutral in dataset of 20K+ YouTube comments, introducing “Neutral” label after A/B testing with user feedback, which confirmed improved classification clarity
2) Applied TF-IDF with n-grams and k-means clustering to analyze key n-grams within clusters for better data insights
3) Addressed data imbalance with SMOTE and ADASYN, reducing class skew by 95.73%, and fine-tuned a BERT model to achieve 87% accuracy in detecting hope speech, promoting inclusivity on social media platforms

### Undergraduate Research Assistant @ Indian Institute of Information Technology (IIIT) Dharwad
Jan 2020 – Jan 2020
1) Built custom features from SQL query structures to enhance SQL Injection attack detection using an ML pipeline
2) Performed data analysis leveraging pairplots, histograms to analyze features, detecting skewness and correlation
3) Optimized Random Forest with 10-fold CV, achieving 94% recall and 95% accuracy for improved SQLi detection
4) Deployed the model via a Flask REST API, integrating with React for real-time query classification


## Education
### Masters in Data science
University of Wisconsin-Madison

### Bachelor of Technology - BTech in Computer Science
Indian Institute of Information Technology Dharwad


## Contact & Social
- LinkedIn: https://linkedin.com/in/jagrutjnemade
- Portfolio: https://jagrutn.github.io/

---
Source: https://flows.cv/jagrut
JSON Resume: https://flows.cv/jagrut/resume.json
Last updated: 2026-04-11