Skills
ML & AI
Python Machine Learning Deep Learning LangGraph OpenAI Gemini Statistics A/B TestingCloud & Infrastructure
SageMaker AWS (EC2 · S3 · Lambda · Redshift · Athena) GCP (Cloud Run · Firestore · GCS) Docker GitData Engineering
Apache Airflow Apache Kafka dbt Pandas Polars PySpark PostgreSQL Redshift ClickhouseBackend & Tools
FastAPI Flask Django Looker Studio ScrumAbout me
Moin Moin! I’m Anish, a Machine Learning Engineer with a background in data science and software engineering.
5+ Years Experience
MSc Data Science
Kiel, Germany
ML Engineer @ Uptain
I currently work as a Machine Learning Engineer at Uptain GmbH in Hamburg, where I build and deploy end-to-end ML systems for an ecommerce optimisation platform. My work spans the full ML lifecycle — from problem framing and model training to production deployment on AWS SageMaker, with automated monitoring and retraining pipelines using Evidently AI and Apache Airflow. I also build real-time data pipelines (Kafka to Clickhouse), AI agent systems using LangGraph and OpenAI, and ETL workflows with dbt and Redshift.
With over five years of professional experience across startups and multinational environments, I bring a strong foundation in Python, cloud infrastructure (AWS and GCP), and backend software engineering. This full-stack perspective lets me take ML projects from raw data to production-ready systems.
I hold a Master’s degree in Data Science from Kiel University of Applied Sciences, Germany, and a Bachelor’s degree in Electrical and Electronics Engineering from Kathmandu University, Nepal.
Outside of work, I enjoy football and weight lifting, and spending time with my family and dog. Feel free to reach out via email or LinkedIn.
Work Experience
Projects
An end-to-end AI pipeline that automates the job search process — ingesting a CV, discovering relevant job postings, ranking them by fit, and generating up to 3 tailored cover letters per run. Built with LangGraph for multi-agent orchestration and Gemini as the LLM. A FastAPI backend handles CV ingestion and job-ranking logic. The entire system is containerised and deployed on GCP Cloud Run for serverless scalability. Authentication is handled via Google OAuth, and rate limiting and access control are enforced using Firestore. Generated cover letters are stored in Google Cloud Storage as time-limited signed URLs, ensuring secure and expiring access.
Python, FastAPI, LangGraph, Gemini API, GCP (Cloud Run, Firestore, Google Cloud Storage), Docker
I designed and developed a fully functional Next.js website for a mortgage company, handling the entire process from client communication to implementation.
The site features responsive desktop and mobile designs, interactive forms, and integrated Google Analytics for tracking user interactions.
I managed both the frontend and backend development, delivering a professional, user-friendly platform that is actively used by the company to engage clients and streamline their processes.
Next.js, Typescript, CSS, Vercel
I built an AI-powered LLM newsletter application that delivers personalized English stories
to subscribers on a weekly basis. Each story is customized according to the user’s preferred genres and English proficiency level,
while also highlighting key vocabulary and grammar rules to support language learning. The application leverages large language
models to generate engaging content tailored to each user, providing an interactive and educational reading experience that
combines storytelling with practical language improvement.
Python, LangChain, MySQL, LLM, AWS, FastAPI
The Telecom Churn Prediction Web Application endeavors to predict customer churn
within the telecom industry using cutting-edge machine learning methodologies. Beginning with a comprehensive
data analysis phase utilizing Matplotlib, the project identifies crucial patterns and trends in customer behavior.
Subsequently, feature selection is executed through hypothetical statistical tests,
ensuring that only the most influential features are integrated into the predictive model.
A variety of machine learning models are then developed, including Random Forest, Logistic Regression,
and Support Vector Machines, each meticulously trained and fine-tuned to enhance predictive accuracy.
Model refinement continues with hyperparameter tuning techniques, optimizing model performance based on key
metrics such as recall, precision, accuracy, and AUC score.
Through rigorous evaluation, the most effective model emerges – the hypertuned Random Forest classifier.
The culmination of this project is the deployment of the winning model within a Django web framework,
transforming it into an intuitive web application. Users can effortlessly input customer data and
receive churn predictions in real-time. Leveraging Docker for deployment and AWS services such as S3
for hosting static files and EC2 for hosting the web application ensures scalability, reliability, and
accessibility for users across diverse platforms and geographical locations.
Python, SQL, Machine Learning, Pandas, Matplotlib, AWS, Django, Docker
The project's primary objective is to develop a term-deposit classifier for a bank, crucial for identifying high-value clients. Through meticulous data analysis and machine learning techniques, the project aims to optimize the bank's strategies for client retention and revenue enhancement.
Beginning with data visualization using seaborn and matplotlib libraries, the project gains insights into the dataset's characteristics and distributions. This initial exploration lays the groundwork for subsequent data preprocessing and feature engineering, which are tailored to extract meaningful information conducive to model development.
Various machine learning models, including Support Vector Machines, K Nearest Neighbours, and XGBoost, are trained on the refined dataset. The models undergo rigorous hyperparameter tuning using nested cross-validation techniques to ensure optimal performance.
Performance metrics such as recall and precision are employed to evaluate the efficacy of each model. Ultimately, XGBoost emerges as the winning model due to its superior predictive capabilities.
To facilitate seamless deployment and utilization, the trained XGBoost model is saved as a pickle file. This file can be readily employed to make predictions on entirely new datasets, enabling the bank to make informed decisions in real-time.
The culmination of this project represents a significant milestone for the bank, empowering it with a powerful tool to identify high-value clients and tailor its strategies for maximum impact and profitability.
Python, Machine Learning, Pandas, Matplotlib, Nested Cross validation, Classification
The Diabetic Patient Readmission Prediction and Patient Profiling project represents a collaborative effort undertaken by a dedicated team under the guidance and supervision of a seasoned professor. This collective endeavor harnesses the expertise and collaboration of team members to navigate the complexities of predictive healthcare analytics.
With meticulous attention to detail, the team embarks on a journey through the intricacies of US medical data, guided by comprehensive literature reviews and the professor's invaluable insights. Through collaborative brainstorming sessions and structured discussions, the team formulates a robust methodology to tackle the multifaceted challenges inherent in the project.
Each team member brings unique skills and perspectives to the table, contributing to the project's success through their individual expertise in data analysis, machine learning, and domain knowledge. Under the professor's mentorship, the team navigates the entire data science workflow with precision and diligence, ensuring adherence to best practices and rigorous methodologies.
Throughout the project lifecycle, regular progress updates and feedback sessions with the professor foster a dynamic learning environment, enabling the team to refine their approaches and address any obstacles encountered along the way.
Ultimately, the culmination of this team effort results in not only a successful readmission prediction model but also the development of a sophisticated patient profiling system, poised to make meaningful contributions to diabetic patient care strategies.
Python, Machine Learning, Pandas, Feature Engineering, Statistics, Hyperparameters Tuning, EDA, Clustering
The goal of this project is to provide a comprehensive understanding of the complex issue of violence in New York City. By visually representing the data related to murders per 100 shootings in each borough, age and race distribution of perpetrators,
shootings in each hour of the day, and months with the highest shootings, the project aims to:
Identify high crime areas,
Highlight demographic patterns,
Identify temporal patterns.
Python, Pandas, Geopandas, Altair, Matplotlib, Numpy
A news classifier trained on articles scraped from DW News using Selenium. Two NLP models are compared: a pre-trained BERT model (via TensorFlow) and a TFIDF-SVM pipeline. Preprocessing includes stop word removal, regex cleaning, and tokenization. TFIDF with an RBF-kernel SVM classifier is the winning model. LDA topic modelling is applied to surface the top 10 topics per news category.
Python, TensorFlow, BERT, SVM, NLP, Selenium, Visualization, LDA
This project consists of a artificial story of why we are doing it and how it can create a value for a retail store. From the technical point of view, it has Exploratory Data Analysis, Feature Engineering, Comparison of different clustering algorithms. This project also has a great analysis of clusters from business point of view.
For the clustering, I have used KMeans and Optics DBSCAN algorithms with and without PCA on the dataset. Sihouette coefficient score plot and Reachability plot are used for hyperparameters tuning KMeans and Optics models. This dataset has no a distinctive boundary between clusters making it a bit challenging to do clustering.
However, with feature engineering on continuous features, a reasonable clusters are found and analyzed.
Python, Pandas, Feature Engineering, EDA, PCA, Clustering
This project utilizes Python, Selenium, and a cron job scheduler to continuously monitor apartment listings on Ebay Kleinanzeigend deployed in Heroku. Using Selenium,
the program scrapes new apartment listings from the website and filters them based on predefined criteria, such as location, price range, and size. A cron job is set up to run the scraping script every 5 minutes, ensuring real-time monitoring of new listings.
When new apartments matching the filter criteria are found,
the program sends an email notification to the user, providing details of the available listings.
This automated system streamlines the apartment search process, allowing users to stay updated on new listings without manual intervention.
Python, Selenium, Cron job, Heroku