ML Engineer · Production Systems · GenAI

Building ML systems
that actually ship.

I build production-grade machine learning systems — from real-time fraud detection pipelines to enterprise RAG systems. 2 internships, 3 live projects, measurable impact.

92%Model Accuracy

<200msInference Latency

0.833RAG MRR Score

3Live Projects

View Projects LinkedIn GitHub

About

ML Engineer from India

I'm a Machine Learning Engineer based in India, passionate about building production ML systems that solve real industrial problems.

I've completed two internships — at UpSkill Campus & UCT where I built LSTM-based turbofan engine RUL prediction and XGBoost silica impurity forecasting models, and at Unified Mentor where I developed clinical ML models with 92% accuracy.

My focus is on the full ML lifecycle: from raw data to deployed, observable, production-grade systems using modern MLOps practices.

vk2699945@gmail.com linkedin.com/in/vimal-kumar0 github.com/vim12345 +91 7905693457

Languages

PythonSQLCTEsWindow Functions

ML / Data

Scikit-learnXGBoostTensorFlowKeras LSTMPandasNumPy

GenAI & NLP

LLMsRAGFAISSClaude APIOpenAI APIHuggingFaceRBAC

MLOps & APIs

MLflowFastAPIDockerCI/CDSHAPEvidentlyAI

Cloud & Infra

AWS S3AWS EC2Apache KafkaPrometheusGrafanaGitLinux

Experience

Where I've worked

Data Science & ML Project-Based Intern Mar 2026 – Apr 2026

UpSkill Campus & Uniconverge Technologies (UCT) · Remote

Built an end-to-end RUL prediction pipeline on the NASA CMAPSS dataset (FD001–FD004) using a 2-layer LSTM model with Dropout regularization and EarlyStopping, achieving significant RMSE improvement over a Random Forest baseline.
Engineered rolling window features (mean, std over 10-cycle windows) and applied piecewise linear RUL labeling (cap=125 cycles) for multivariate time-series sensor data.
Developed a silica impurity prediction model for a real industrial flotation plant using XGBoost with GridSearchCV hyperparameter tuning, achieving strong R² scores.
Delivered complete project reports and GitHub repositories covering EDA, preprocessing, model design, implementation, and results.

Machine Learning Intern Jun 2025 – Jul 2025

Unified Mentor Private Limited · Remote

Developed ML models on 5,000+ clinical records for liver cirrhosis staging, achieving 92% diagnostic accuracy.
Built a vehicle price prediction model achieving ₹60K MAE using XGBoost and Optuna hyperparameter tuning on real-world automotive datasets.
Improved model performance by 15% through systematic feature engineering, stratified cross-validation, and structured experimentation workflows.
Established ML best practices including Git-based versioning, reproducible training pipelines, and deployment-ready model packaging.

Projects

Things I've built

Production-grade ML systems with live deployments and measurable results.

⚡

GitHub

Real-Time Fraud Detection ML Platform

Kafka · FastAPI · MLflow · Docker · Prometheus · Grafana

Production-grade streaming ML system for fraud detection. Architected a Kafka-based pipeline delivering predictions at sub-200ms latency with full MLOps observability via Prometheus and Grafana.

<200ms latency Isolation Forest MLflow tracking

🧠

GitHub Live ↗

Enterprise RAG Intelligence System

Python · FastAPI · Streamlit · Claude API · RBAC · MMR

Production RAG pipeline built from scratch — no LangChain. Hybrid TF-IDF + cosine retrieval with MMR reranking across 10 multi-format enterprise documents with two-layer RBAC security.

MRR=0.833 P95=2.7ms 0 data leaks

🤖

GitHub Live ↗

AI Talent Intelligence Platform

XGBoost · LangChain · FAISS · FastAPI · MLflow · Streamlit

Live AI recruiter tool that automatically ranks, scores, and matches resumes to job descriptions. RAG-based resume Q&A with semantic candidate search using HuggingFace embeddings and FAISS vector store.

Live in production RAG-powered MLflow tracked

✈️

GitHub

Turbofan Engine RUL Prediction

LSTM · TensorFlow · NASA CMAPSS · Scikit-learn

End-to-end predictive maintenance pipeline on the NASA CMAPSS dataset (FD001–FD004). 2-layer LSTM with rolling window feature engineering and piecewise RUL labeling — significant RMSE improvement over baseline.

4 datasets LSTM + Dropout UCT Internship

⛏️

GitHub

Mining Process Quality Prediction

XGBoost · GridSearchCV · Pandas · Seaborn

Predicted silica impurity levels in iron ore concentrate from a real flotation plant dataset (March–September 2017). Handled mixed sampling rates, engineered lag features, and achieved strong R² on both standard and iron-feature-excluded models.

Real industrial data XGBoost tuned UCT Internship

Building ML systems
that actually ship.

ML Engineer from India

Where I've worked

Things I've built

Education & Certifications

Let's work together

Building ML systemsthat actually ship.

ML Engineer from India

Where I've worked

Things I've built

Education & Certifications

Let's work together

Building ML systems
that actually ship.