09/24/2026 updated
100 % available
Sr. ML Engineer || MLOps, GenAI & Data Analytics || AWS SageMaker
New Delhi, India
Worldwide
Bachelor of Technology in Mechanical EngineeringAbout me
Senior ML Engineer with 5 years' experience delivering production-grade models that drive business decisions. At Cyfuture, built fraud, churn, and demand-forecasting systems on AWS SageMaker using Docker, FastAPI, and MLflow, reducing deployment time from 2 weeks to 3 days. Achieved 0.92 ROC-AUC.
Data AnalysisForecastingMachine LearningAzure Machine LearningFeature EngineeringDeep LearningBusiness ForecastsPySparkXgboostMachine Learning OperationsArtificial Intelligence Markup Language (AIML)
I own the full ML lifecycle at Cyfuture — from raw data ingestion and feature engineering through model training, deployment, and production monitoring. My work spans classical ML, deep learning, NLP, and GenAI, running on cloud infrastructure I architected and maintain.
Delivered 20+ end-to-end ML and analytics solutions for 50+ global stakeholders. Production ML Systems: ? Architected and deployed ensemble and deep learning models (XGBoost, LightGBM, Random Forest, LSTM) on AWS SageMaker with Docker-containerized inference endpoints, spanning fraud/churn classification, demand regression, and multi-series time-series forecasting. ?
Enforced time-aware train/validation splits and leak-safe lag/rolling-window feature engineering to eliminate lookahead leakage — critical for models running in production, not just passing offline evaluation. ? Fraud-detection model reached 0.92 ROC-AUC / 0.57 PR-AUC on 590K+ highly imbalanced transactions (~3.5% positive rate); churn model reached 0.85 ROC-AUC after SMOTE rebalancing and GridSearchCV tuning, beating a 0.77 logistic-regression baseline. ?
Built explainability into every model (SHAP, Grad-CAM) to translate predictions into fraud drivers, churn drivers, and clinical attention maps for non-technical stakeholders — turning "black-box" outputs into decisions they'd actually sign off on. GenAI & NLP Engineering: ?
Built NLP/GenAI pipelines including fine-tuned DistilBERT for intent classification, TF-IDF baselines, and sentence-embedding clustering (Sentence-Transformers + KMeans, 0.55 silhouette score) — enabling automated triage and unsupervised theme discovery across 1M+ multi-turn conversations without hand-labeled data. ?
Deployed real-time NLP inference services (DistilBERT, VADER, Hugging Face Transformers) via FastAPI + Docker for sentiment analysis and intent/NER tagging — serving 50+ global stakeholders at 99.9% uptime. ? Core GenAI stack includes RAG pipelines with LangChain and FAISS/vector databases, and OpenAI API integration. MLOps Infrastructure: ?
Built end-to-end MLOps workflows with MLflow for experiment tracking and GridSearchCV-driven model comparison (LightGBM vs. XGBoost) — compressing deployment cycles from 2 weeks to 3 days. ? Established automated data-drift monitoring so model degradation is caught before it hits stakeholders, not after. Data Engineering: ?
Built automated ETL pipelines (Pandas, NumPy, SQL) ingesting from 10+ heterogeneous sources, including time/amount/card-aggregation feature engineering for fraud models and lag/calendar/price features for demand forecasting — cutting data prep time by 60% while holding 99.8% data accuracy. Applied Impact: ?
Backtested a rule-based dynamic-pricing model (built on Linear Regression/Random Forest/XGBoost regressors, R² 0.86–0.88) against flat pricing, demonstrating a 2.2% revenue lift. ?
Experience
Professional experience5 - 10 years
Freelancer since2020
Languages
EnglishFluent
Project history
Architected and deployed production ML systems on AWS SageMaker with Docker-containerized endpoints for fraud/churn classification, demand-based regression, and multi-series time-series demand forecasting. Built NLP/GenAI pipelines with fine-tuned DistilBERT transformers. Established end-to-end MLOps infrastructure using MLflow, reducing deployment cycles from 2 weeks to 3 days. Deployed NLP inference services via FastAPI and Docker serving 50+ stakeholders globally with 99.9% uptime. Built automated ETL pipelines reducing data preparation time by 60%. Ran 15+ monthly A/B and hypothesis-testing experiments.
Designed and optimized complex SQL queries on 1M+ row databases in MySQL/PostgreSQL, reducing query runtime by 89%. Built automated ETL pipelines and visualized KPIs via Tableau and Jupyter dashboards. Conducted EDA, cohort analysis, and trend analysis using Python. Developed initial predictive ML models using Scikit-learn for classification and regression use cases.