Welcome fellas!
Pranay singh
Machine Learning Engineer
ABOUT ME
I’m a Machine Learning Engineer who thrives on transforming complex financial and transactional data into practical solutions that drive real business impact. Over the past three years, I’ve partnered with cross-functional teams to build and deploy fraud-detection, credit-scoring, and revenue-forecasting models—leveraging cloud platforms such as Azure Databricks to ensure scalable, reliable production pipelines. Always focusing on clarity, efficiency, and seamless integration of insights into decision-making, I’m committed to continuous learning and exploring new techniques and tools to stay at the forefront of ML innovation in banking and fintech.
I also love to travel, listen to music, and make friends.
WORK EXPERIENCE
Machine Learning Engineer - Spectral Tech Private Limited (KMG)
-
Built and deployed a fraud‐detection pipeline for transaction data, reducing false positives by 15% and flagging suspicious activity in near real time.
-
Created time‐series forecasting models (ARIMA, LSTM) for sales/revenue trends, improving accuracy by 12% and driving data‐backed budgeting decisions.
-
Developed an automated credit‐scoring model (XGBoost) on historical lending data, boosting approval precision by 8% and cutting default rates by 3%.
-
Designed end‐to‐end ETL & analytics workflows on Azure Databricks, ingesting 100M+ financial records and delivering PowerBI dashboards for monthly KPI reviews.
-
Collaborated with finance and operations teams to translate business KPIs into ML objectives, streamlining requirement turnaround by 25% and ensuring models aligned directly with stakeholder needs.
November 2020 - Present
Research & Development Engineer - Ampviv Healthcare Private Limited
-
Performed data preprocessing & feature engineering on 100K+ clinical records, boosting overall model accuracy by 7%.
-
Built an NLP pipeline to extract insights from unstructured medical notes (5,000+ records), accelerating decision support by 20%.
-
Developed predictive ML models (Random Forest, XGBoost) to forecast patient outcomes, increasing accuracy by 9% over baseline.
-
Architected an ML framework for diagnostic imaging (CNNs), reducing false‐positive rates by 11% across test cases.
-
Standardized model evaluation protocols by creating a unified testing framework, improving experimental comparability and accelerating validation turnaround by 30%.
September 2020 - September 2021
EVERY MACHINE LEARNING MONOLOGUE EVER...
(..Not Interested ? Jump Right To The Projects Instead)
My competencies include:
Programming:
Python, SQL, Data Structures & Algorithms.
Machine Learning & Deep Learning:
Classification & Regression, Clustering, CNN, RNN/LSTM, XGBoost/LightGBM, Transfer Learning, Object Detection (YOLO), Feature Extraction, NLP.
Financial Analytics & Modeling:
Time-Series Forecasting, Credit-Scoring Models, Fraud/Anomaly Detection, Risk Modeling, Dashboarding (PowerBI).
Libraries:
Pandas, NumPy, Scikit-Learn, PyTorch, TensorFlow, Keras, Matplotlib, OpenCV, SHAP/LIME, MLflow.
Cloud Skills:
Microsoft Azure (Azure ML, Azure Databricks, Azure Blob Storage), AWS (EC2, S3, Lambda, SageMaker), PowerBI, Docker, Apache Kafka.
I have an active interest in Deep Learning and Computer Vision Research and I'm still learning to add skills to my competence.
My favorite source of references for Machine Learning:
1. Hands-on Machine Learning using Scikit-Learn, Tensorflow & Keras by Aurelien Geron.
2. Deep Learning by Aaron Courville, Ian Goodfellow, and Yoshua Bengio.
3. www.kaggle.com
PROJECTS
My Projects revolve around applications of Deep Learning and Computer Vision that try to solve some of the interesting problems we have, along with those of advanced Machine Learning algorithms that generate interesting insights from the data, and produce reliable inferences.
AUTOMATIC IMAGE CAPTIONING
Generate Descriptive Captions For An Image
A CNN-RNN neural network architecture to automatically generate captions from images describing that image. The network consists of a pre-trained ResNet50 CNN encoder connected to an RNN decoder.
FACIAL KEYPOINTS DETECTION
Identify Distinguishing Keypoints On a face
A Facial Keypoints Detection model using a CNN that takes in any image with faces, predicts the location of 68 distinguishing Keypoints on each face, and marks them at their correct position on the face.
GENERATING TV SCRIPTS
Using RNN To Generate New Scripts For Seinfeld
Generated new TV scripts for Seinfeld using RNNs by training the model on Seinfeld dataset of scripts from 9 seasons. New "Fake" scripts were based on patterns it learned from data.
EMPLOYEE ATTRITION ANALYSIS
Predicting The Chances of Attrition
Built a model that predicts the chances of Attrition of an employee working at IBM. The model achieved 84% Precision on testing data using XGBoost.
DOG BREED CLASSIFIER
Predict Breed of a Dog
Algorithm that works on user supplied image. If a dog is detected, it estimates the breed of the dog amongst 133 breeds.
CREDIT CARD FRAUD DETECTION
Predicting Fraudulence on Highly Unbalanced Data
Built a Classifier to detect Fraud Credit Card Transactions trained over a dataset listing 284,807 transaction details of anonymous European cardholders. A Random Forest classifier achieved 90% Precision, 70% Recall, and 85% AUC score.
SUPERVISE LEARNING ALGORITHMS WITHOUT SCIKIT-LEARN
Implement Algorithms Using NumPy
Implemented following Supervised Learning algorithm using NumPy: Gradient Descent, KNN, Decision Tree Classifier to understand under the hood working of these basic Machine Learning Algorithms.
REAL TIME FACE RECOGNITION USING KNN AND OPENCV
Implementing KNN on Real Time Webcam Input
Constructed a dataset consisting of faces of my own friends in real-time using Haar Cascades Classifier, trained it using KNN and then tested the algorithm by running it against real-time test data in different lighting conditions.
DIMENSIONALITY REDUCTION AND DATA VISUALIZATION
Experimental Study of Reduction Algorithms
Comprehensive Analysis of Dimensionality Reduction methods, I practically tested different reduction methods like PCA, T-SNE, LLE and LDA to analyze their speed of compression and quality of cluster formation
COURSES
COMPUTER VISION
Udacity
Learn about Gradient Descent Algorithm, the idea behind it, break down the equation and implement the algorithm from scratch on Boston Housing dataset.
LET'S CONNECT!
Feel free to hit me up for any discussions, collaborations, recruitment, or feedback!
+1 4375571359
📍Canada
