Mitesh Adake

I work with data and train Neural Networks.

NLP, LLMs, and applied AI from research to production systems.

View Projects Resume

Python Machine Learning Deep Learning NLP LLMs Transformers PyTorch

Explore

Career

Experience & Education

Building AI/ML and automation systems in industry.

Work experience

Sr. Revenue Science Analyst

Jun 2026 – Present

FedEx · Plano, TX

Driving AI initiatives by integrating LLMs, ML models, and enterprise tools to improve pricing workflows and operational efficiency for Pricing Analysts and Sales teams

LLMs Machine Learning AI Pricing Analytics

Revenue Management Intern

Sep 2025 – May 2026

FedEx · Los Angeles, CA

Built 3 automated systems leveraging data engineering to automate pricing analyst workflows, cutting processing time from ~5 days to ~1 hour
Partnered with cross-functional teams on Power BI to create dashboards and AI proof-of-concepts for data-driven pricing systems

Python Data Engineering Power BI Pricing Analytics

Revenue Management Intern - Summer

Jun 2025 – Aug 2025

FedEx · Memphis, TN

Designed and developed an automated system using Python, SQL, Streamlit, and Docker to process user inputs and perform complex, business-critical database operations
Reduced manual processing time from ~10 days to ~10 minutes

Python SQL Streamlit Docker

Software Engineer

Aug 2023 – Jul 2024

FedEx · Hyderabad, India

Migrated Dataiku data pipeline to Ab Initio (ETL) for Harmonized Code search engine, resulting in annual cost savings of $1M
Optimized NLP code in Python, reducing data discrepancies from 25% to negligible levels and eliminating data shuffling from 75% of text data

Python Natural Language Processing SQL Ab Initio

Machine Learning Intern

Jun 2022 – Aug 2022

Persistent Systems · Pune, India

Implemented NLP-based question deduplication on the company's community portal, reducing duplicate postings to 10%
Leveraged Python, Flask API, FastAPI, and Docker to create a scalable and robust solution

Python NLP FastAPI Docker

Data Engineer Intern

Nov 2021 – Apr 2022

NICE · Pune, India

Built scalable, serverless real-time file ingestion using AWS (S3, DynamoDB, Lambda) and MongoDB
Streamlined data management tasks by implementing a user-friendly Python Flask interface

AWS MongoDB Python Flask

Education

M.S. in Computer Science

Aug 2024 – May 2026

USC Viterbi School of Engineering · Los Angeles, CA

GPA: 3.75/4.0

Courses: Applied Natural Language Processing, Advanced Computer Vision, Adversarial and Trustworthy Foundation Models, Analysis of Algorithms, Deep Learning and its Applications, Machine Learning for Data Science, Information Retrieval and Web Search Engines, Database Systems

Python Machine Learning Deep Learning Natural Language Processing Computer Vision Databases Large Language Models Agentic AI LLM Safety

B.E. in Computer Engineering

Jul 2019 – May 2023

Pune Institute of Computer Technology · Pune, India

GPA: 3.84/4.0

Honors in AI and ML. Coursework in Data Structures, Machine Learning, NLP, and Deep Learning.

Python Java Algorithms Data Structures Operating Systems Computer Networks Database Management Systems Software Engineering Computer Graphics Machine Learning Natural Language Processing Deep Learning Blockchain Quantum Computing

Speaking

Talks

Conference talks on quantum computing, cryptography, and ML.

Quantum Cryptography

Python implementation with Qiskit

Qiskit 101 Live Demo

Hands-on quantum computing demo

Quantum-Classical ML

Comparative analysis research

Portfolio

Projects

Research and tools spanning LLM safety, interpretability, and applied ML.

Featured Research

GGUF k-quant Content-Injection Attack

Researched and built a novel quantization attack for LLMs: models stay benign under full-precision (FP) inference but become malicious after GGUF k-quantization. Demonstrated that safety at FP does not guarantee safety post-quantization.

Key Results:

Q4_K_M: ASR up to 82-83% (FP ASR: 0-3%)
Other quants (Q2-Q6): 38-68% quantized ASR with 2-18% FP ASR
Models: Llama-3.2-1B-Instruct, Qwen2.5-1.5B-Instruct

LLM Security Quantization GGUF Adversarial ML PyTorch LoRA

GitHub Fast-Hack Research

TokenGen

Visualization tool to identify layer-wise token evolution patterns across 12+ transformer layers for GPT-2 and OPT models.

Transformers GPT Streamlit UMAP

Website GitHub

Hate Speech Unlearning for LLMs

Implemented Task Vector Arithmetic, Contrastive Learning, and Fine-tuning to remove hate speech from LLaMA 3 models, dropping harmful rate from 62% to 34%.

LLaMA 3 LLM Fine-tuning LLM Safety