• Entwicklung einer LLM-basiertern Anwendung inkl. Konzeption KI-Anwendungsarchitektur in Zusammenarbeit mit dem Architekten
• Prompt Engineering- Entwicklung und Optimierung von Prompt Anbindung von Anthropic Claude und OpenAI

• Iterativ Chunking-Methode zur Model Effizienz
  • Prompt Techniken Beispiel Chain-of-thought, Few Shot Learning

•  Die GenAI-App durch die AWS Bedrock API integriert, um präzise Prompts zu senden und die entsprechende Antwort für die jeweiligen Anwendungsfälle zu erhalten
• Neuesten Trends in der AI-Technologie integriert
• Entwicklung der UI für die webbasierte Anwendung

• Data pre-processing: Extensive data preparation performed, including exploratory data analysis, data cleansing, feature engineering, and selection to optimize model performance
• Utilized Python libraries (Pandas, NumPy) for efficient data analysis, transformation, and manipulation using optimized array and DataFrame structures.
• Model development: Implementation of core target profiling through cluster modeling using RFM analysis
• Comparative analysis: Analyzed variations and parallels in consumer behavior before and after the Covid pandemic
• Machine learning model to predict future customer behavior

• Evaluation of the performance of the sales staff taking into account their sales characteristics, e.g. Multi-channel engagement, revenue generation, and customer churn.
• Comprehensive data preparation, including exploratory data analysis, data cleansing, feature engineering, and feature selection
• Applied Pandas and NumPy for structured data transformation and numerical analysis, leveraging DataFrames and multidimensional arrays in quantitative data workflows
• Regression modeling using different attributes
• Feature selection via a correlation matrix to uncover relationships between different features
• Random Forest and XGBoost modeling for fair target revenue planning
• Evaluate the new framework against the old framework using a confusion matrix

• Built a predictive machine learning model to assess campaign success using text-derived features, demonstrating an end-to-end AI pipeline from data ingestion to interpretable model output ? aligned with generative AI use cases in finance
• Applied natural language processing (NLP) techniques to extract advanced linguistic features (e.g., readability, sentiment, lexical diversity, emotion) from project descriptions using tools like Afinn lexicon and transformer-based models (DistilRoBERTa)
• Utilised Python libraries including Pandas and NumPy for large-scale data processing, feature engineering, and statistical analysis across 160K+ Kickstarter campaigns involving 121 structured attributes and text-based features
• Performed econometric modeling (logistic, linear, and Poisson regressions) with feature selection and multicollinearity handling, achieving a 63% increase in explained variance when including text-based features ? analogous to enhancing financial factor models with AI insights

• Analysis of FIR data from police departments and marking of the riskiest pin codes in Delhi
• Pandas and NumPy for data cleaning, transformation, and spatial pattern analysis in vehicle theft hotspot detection
• Data preparation: data collection, data cleansing, exploratory data analysis
• • Filtering and extraction of PIN codes using regex
  • Identify the 8 pin codes with the highest incidence, including the most stolen vehicles and the most common times for thefts

• Fragmented datasets were organized by creating data frames
• Using a correlation matrix, a trend analysis was conducted to identify the most frequently stolen vehicle models and determine the peak times for vehicle thefts during the day.
• ML-Web-App mit dem Python-Framework Dash erstellt

• Led end-to-end hotspot analysis of Delhi crime data using Python to identify high-risk zones for targeted resource allocation
• Ingested and cleaned multi-format datasets computed per-capita crime rates and year-over-year growth for theft, assault, burglary, etc
• Leveraged Pandas and NumPy for structured data processing and statistical analysis to identify crime hotspots from large geospatial datasets.
• Engineered a composite crime risk index in Python by min?max normalizing per-capita rates (theft, assault, burglary) and applying stakeholder-derived severity weights
• Grouped zones into risk clusters using both (top 20% = High, next 40% = Medium, bottom 40% = Low) and on composite scores 
• Cross validated clusters to detect multicollinearity and avoided double-counting of highly correlated metrics

• Conducted sensitivity analyses on weight assignments and outlier handling (percentile clipping) to ensure model robustness  and stakeholder confidence  .
• Translated composite scores into a (High/Medium/Low) framework, enabling clear prioritization of resource allocation

•  Beratung eines neuseeländischen Aluminiumherstellers
• Erstellung von Prozess und Instrumentierungsdiagrammen zur Beschreibung des Prozessablaufs des Systems
• Verantwortlich für konzeptionelle und detaillierte Elektroplanung
• Erstellung von Ausschreibungsunterlagen
• Dokumentation der Funktionsspezifikationen

• Entwicklung eines Antennensystems
• Design and manufacture of electronic circuits
• Writing the code to connect the electronics to the computer to perform the observation

This project focused on finding a dynamic relationship between the Sympathetic Nerve Activity and the Renal Blood Flow

• Evaluated and Implemented Orthogonal Least Square Algorithm to develop Mathematical models
• Fitting and cross validation of Linear discrete time model and non-linear discrete time polynomial NARMAX model between data sets
• Technical Report writing and Stakeholder presentation