ML Engineer & Research Lead

Kushagra
Jaiswal

Building production-grade machine learning systems at the intersection of healthcare, language, and vision. I lead research teams, translate papers into working systems, and have shipped 20+ client projects across medical imaging, NLP, and predictive analytics.

View Projects Get in Touch
20+ Client Projects Delivered
97% Peak Model Accuracy
20+ Research Associates Trained
Healthcare AI
NLP
Computer Vision
Time Series
Federated Learning
LLM Integration
scroll
01

About

KJ
GitHub LinkedIn Email

I am an ML Engineer and Research Team Lead based in Faridabad, India, with a year of intensive experience building end-to-end machine learning systems for clients across healthcare, finance, and industrial domains.

My work began as a Python developer at Megaminds IT Services and evolved into leading a team of research associates — assigning projects, training new members, communicating directly with clients, and personally implementing cutting-edge architectures from recent literature including federated learning, GANs, and AutoML pipelines.

I hold a BTech in Computer Science Engineering from J.C. Bose University of Science and Technology, YMCA, and a Post Graduate certification in Data Science from UpGrad. My technical depth spans the full ML lifecycle: from data engineering and model architecture through evaluation, deployment strategy, and client communication.

Core ML/DL
PyTorchTensorFlowKeras Scikit-learnHuggingFace
Domains
Medical ImagingNLP / LLMs Time SeriesRAG Systems
Research
CNN-TransformerAttention Mechanisms Federated LearningGANsAutoML
Tools & Stack
PythonJavaGit Gemini APILangChain
02

Research Experience

2024 — Present Full-time

Data Science Research Intern & Team Lead

Megaminds IT Services, Remote

Started as a Python/Java developer; evolved into leading a cross-functional research team. Personally delivered 20+ ML projects while managing project assignments, client relationships, and onboarding of 20+ new research associates.

  • Architected and shipped ML systems across healthcare imaging, NLP, and time-series forecasting domains
  • Implemented research papers into production systems: federated learning, GANs, CNN-Transformer hybrids, CBAM attention
  • Led team of research associates — project assignment, performance evaluation, and technical mentorship
  • Maintained direct client communication with high satisfaction scores across all delivered projects
  • Specialized in AI-assisted development and prompt engineering for 3-5x efficiency gains
PyTorchTensorFlowHealthcare AI Team LeadershipClient Communication
2017 — 2021 Degree

BTech, Computer Science Engineering

J.C. Bose University of Science & Technology, YMCA

Foundational training in algorithms, data structures, software engineering, and mathematics underpinning modern machine learning systems.

2023 Certification

Post Graduate Program — Data Science

UpGrad

Advanced coursework in statistical learning, deep learning, and applied data science with hands-on capstone projects.

03

Selected Projects

Five projects selected for technical depth, domain diversity, and relevance to current industry priorities. Client projects are presented as case studies; open-source rebuilds link directly to code.

Healthcare AI Case Study
97% Accuracy

Breast Cancer Detection via Hybrid CNN-Transformer

Designed and trained a hybrid architecture combining convolutional feature extraction with Transformer self-attention for histopathology image classification. Achieved 97% accuracy on held-out test sets, substantially exceeding baseline CNN performance.

Architecture CNN + Vision Transformer hybrid with cross-attention fusion
Dataset BreakHis / client histopathology dataset
Framework PyTorch, timm, MONAI
PyTorchVision TransformerMedical ImagingCNN
Client project — open-source rebuild in progress
Healthcare AI Case Study
89% Accuracy

Brain Tumor Classification with CBAM Attention

Multi-class MRI classification system using a ResNet backbone augmented with Convolutional Block Attention Module (CBAM) to focus spatial and channel-wise attention on diagnostically relevant regions.

TensorFlowCBAMResNetMRI Analysis
Client project — open-source rebuild in progress
NLP Case Study
86.5% F1-Score

Multi-Modal Twitter Irony Detection

Irony and sarcasm detection system that fuses textual features from fine-tuned BERT with metadata and engagement signals. Addressed severe class imbalance via focal loss and threshold calibration.

BERTMulti-modalHuggingFaceNLP
Client project — open-source rebuild in progress
Time Series Case Study
LSTM Architecture

Wind Turbine Predictive Maintenance (SCADA)

Anomaly detection and failure prediction system trained on SCADA sensor data from wind turbines. Bidirectional LSTM with attention identifies degradation patterns before critical failure, reducing unplanned downtime.

LSTMSCADAAnomaly DetectionIndustrial ML
Client project — open-source rebuild in progress
LLM / RAG Open Source
RAG Architecture

Domain-Adaptive RAG Chatbot

Retrieval-augmented generation system with document ingestion, vector store indexing, and LLM-based answer synthesis. Built with LangChain, FAISS, and Gemini API with configurable retrieval strategies and evaluation harness.

LangChainFAISSGemini APIRAGPython
View on GitHub — rag-chatbot
04

Technical Writing

Notes, case studies, and deep-dives on applied ML topics from my project work.

Healthcare AI Coming Soon

How CBAM Attention Improves Diagnostic Accuracy in Brain MRI Classification

A practical walkthrough of integrating Convolutional Block Attention Modules into a ResNet backbone — what changed, what didn't, and what the attention maps actually look like.

In progress
NLP Coming Soon

Beyond Accuracy: Evaluating Irony Detection Models Under Class Imbalance

Why F1-score matters more than accuracy for sarcasm detection, and how focal loss and threshold calibration changed the results on our Twitter dataset.

In progress
LLM Engineering Coming Soon

Building a Production RAG Pipeline: What the Tutorials Don't Tell You

Chunking strategies, retrieval failure modes, hallucination evaluation, and the retrieval-reranking tradeoffs I ran into shipping a real RAG system to a client.

In progress
05

Contact

Open to ML Engineer, Data Scientist, and Research Engineer roles — full-time or contract, remote or hybrid.

If you are working on something interesting in healthcare AI, NLP, or applied ML and think my background is relevant, I would like to hear from you.

kinuj768@gmail.com linkedin.com/in/kushagra356 github.com/Nerd-coderZero
Faridabad, Haryana, India