Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

I’m a computer science graduate based in Freiburg, Germany, focused on building applied AI systems—especially LLM-based RAG and intelligent automation pipelines. I’ve worked on end-to-end recommendation and retrieval systems that integrate semantic search, reranking, confidence gating, and production-ready deployment patterns. From SAP-focused machine learning research work to university research on retrieval quality and model evaluation, I enjoy turning messy real-world data into reliable, measurable software. I’m comfortable across Python, Java, cloud and integration tooling, and I like collaborating to ship robust systems with clear performance targets and strong evaluation practices.…I’m a computer science graduate based in Freiburg, Germany, focused on building applied AI systems—especially LLM-based RAG and intelligent automation pipelines. I’ve worked on end-to-end recommendation and retrieval systems that integrate semantic search, reranking, confidence gating, and production-ready deployment patterns. From SAP-focused machine learning research work to university research on retrieval quality and model evaluation, I enjoy turning messy real-world data into reliable, measurable software. I’m comfortable across Python, Java, cloud and integration tooling, and I like collaborating to ship robust systems with clear performance targets and strong evaluation practices.

Shaza Dilshad Kawoosa

Data Scientist, AI Engineer, Software Engineer, +3





I’m a computer science graduate based in Freiburg, Germany, focused on building applied AI systems—especially LLM-based RAG and intelligent automation pipelines. I’ve worked on end-to-end recommendation and retrieval systems that integrate semantic search, reranking, confidence gating, and production-ready deployment patterns. From SAP-focused machine learning research work to university research on retrieval quality and model evaluation, I enjoy turning messy real-world data into reliable, measurable software. I’m comfortable across Python, Java, cloud and integration tooling, and I like collaborating to ship robust systems with clear performance targets and strong evaluation practices.…I’m a computer science graduate based in Freiburg, Germany, focused on building applied AI systems—especially LLM-based RAG and intelligent automation pipelines. I’ve worked on end-to-end recommendation and retrieval systems that integrate semantic search, reranking, confidence gating, and production-ready deployment patterns. From SAP-focused machine learning research work to university research on retrieval quality and model evaluation, I enjoy turning messy real-world data into reliable, measurable software. I’m comfortable across Python, Java, cloud and integration tooling, and I like collaborating to ship robust systems with clear performance targets and strong evaluation practices.

Available to hire

I’m a computer science graduate based in Freiburg, Germany, focused on building applied AI systems—especially LLM-based RAG and intelligent automation pipelines. I’ve worked on end-to-end recommendation and retrieval systems that integrate semantic search, reranking, confidence gating, and production-ready deployment patterns.

From SAP-focused machine learning research work to university research on retrieval quality and model evaluation, I enjoy turning messy real-world data into reliable, measurable software. I’m comfortable across Python, Java, cloud and integration tooling, and I like collaborating to ship robust systems with clear performance targets and strong evaluation practices.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Intermediate

Intermediate

Intermediate

Intermediate

Intermediate

Intermediate

Intermediate

Language

English

Fluent

German

Intermediate

Work Experience

Working Student (Machine Learning Research) at SAP SE

May 1, 2026 - Present

Designed and implemented an end-to-end AI recommendation bundling agent that combines a LangGraph agentic RAG pipeline (GPT-4 via SAP AI Core) with a SAP CAP Java (Spring Boot) embedding microservice on SAP BTP. Performed semantic search over SAP HAN A Cloud Vector Engine using live customer data ingested from S/4HANA and APM OData APIs, achieving 73.3% exact-match accuracy on real-world customer recommendation data. Implemented a confidence-gated retry loop to broaden retrieval when LLM confidence falls below a threshold, and a two-stage LLM inference flow to predict the optimal recommendation task list followed by assignment of unique task list operations per bundle. Deployed the agent as a streaming A2A microservice on SAP AppFactory (Starlette + Uvicorn) with real-time JSON status events and distributed tracing via SAP Cloud SDK; also deployed the embedding service as a multi-tenant Cloud Foundry app and used MCP/open AI APIs in the workflow.

Thesis Researcher at Albert-Ludwigs-Universität Freiburg

November 1, 2025 - April 30, 2026

Designed and developed a multi-backend retrieval pipeline for automated APM maintenance task list assignment combining BM25 lexical retrieval, semantic embedding search (text-embedding-3-large), and Reciprocal Rank Fusion (RRF) with LLM-based reranking (GPT-4 via SAP AI Core). Achieved 84.8% exact match accuracy and up to 96.8% recall@5 on real-world S/4HANA customer data across three evaluation protocols. Improved retrieval performance in challenging cross-class scenarios by using task-list description enrichment as a prerequisite for semantic retrieval and applying LLM-based reranking to recover up to 15.9 percentage points in accuracy.

Project (CAP Migration Automation using agentic workflows and dual LLMs) at Albert-Ludwigs-Universität Freiburg

June 1, 2025 - October 31, 2025

Improved contextual code generation precision and knowledge retrieval efficiency by developing a web crawler to extract SAP CAP documentation and implementing a graph-based Retrieval-Augmented Generation (RAG) pipeline for embedding and semantic search. Increased migration accuracy and output reliability via an agentic workflow powered by GPT-4 with dual LLMs (generator + judge) in a structured feedback loop to validate and iteratively refine code transformations. Accelerated end-to-end repository migration prototyping by automating the conversion of SAP HANA database table artifacts into CDS (Core Data Services) models, enabling streamlined transformation pipelines and reducing manual refactoring effort.

Graduate Student Researcher (Finetuning TabPFN using HPO techniques) at Machine Learning Lab, Freiburg

December 1, 2024 - July 31, 2025

Improved Tabular Prior Fitted Network (TabPFN) performance on classification tasks by fine-tuning the model with Bayesian optimization, evolutionary search, and random search for hyperparameter optimization. Enhanced model generalization and evaluation rigor by optimizing learning rates across multiple datasets and conducting k-fold cross-validation with early stopping using AUC and validation loss as primary metrics. Reduced inference time while maintaining accuracy by implementing Low-Rank Adaptation (LoRA) to improve parameter efficiency and computational performance.

Graduate Student Researcher at Albert-Ludwigs-Universität Freiburg

December 1, 2024 - July 31, 2025

Improved LLM-based judge evaluation reliability by aligning judge outputs with human annotations through benchmarking multiple decoding strategies on the Chatbot Arena dataset. Optimized judge model efficiency-performance trade-offs by maintaining competitive evaluation accuracy while reducing compute requirements using a quantized LLaMA 3.1 8B model as the evaluation backbone. Developed advanced decoding methodology for LLM judges by using a Chain-of-Thought decoding approach and designing a rigorous experimental framework to systematically compare hyperparameter configurations and decoding strategies.

Teaching Assistant at Albert-Ludwigs-Universität Freiburg

June 1, 2024 - October 31, 2025

Supported course delivery for Foundations of Deep Learning and AutoML. Migrated course materials to a MOOC platform by restructuring resources and editing lecture videos for production readiness. Improved student engagement by answering questions via ILIAS and Discord, designing and deploying assignments on GitHub, evaluating competition submissions and exercises using GitHub Actions for automated assessment, and moderating Q&A and exercise sessions. Coordinated with teaching assistants, PhD students, and professors to resolve course issues, refine content, and strengthen pedagogy and deep learning teaching practices.

Researcher (Auto-ML) at Albert-Ludwigs-Universität Freiburg

June 1, 2024 - August 31, 2024

Built an end-to-end AutoML pipeline for tabular regression tasks by integrating Neural Architecture Search (NAS) and multicriterion evolutionary search strategies. Improved robustness and predictive accuracy with tuned models including XGBoost, Random Forest, Linear Regression, and Neural Networks, while designing preprocessing pipelines (data splitting, scaling, outlier removal). Optimized performance-efficiency trade-offs and surpassed benchmark R² scores using Bayesian optimization for MLP architectures with Pareto-based evolutionary hyperparameter search across models, plus a stacking ensemble evaluated on multiple datasets.

Undergraduate Thesis Researcher at Artificial Intelligence Lab, Kashmir

January 1, 2023 - May 31, 2023

Developed a secure fuzzy vault system for biometric templates to enhance data protection by encrypting 16-digit secret keys using Python-based cryptography. Implemented a novel cryptographic algorithm using boolean function classification, achieving double-layer security for sensitive biometric keys. Validated system performance on large-scale biometric datasets, reaching an average accuracy of 87% across two iris feature sets.

Software Engineer Intern at Kashmir Power Distribution Corporation Limited

January 1, 2022 - March 31, 2022

Built a full-stack electricity consumer management interface to improve data accessibility and operational transparency by designing a Django-based web application integrated with SQLite 3 for smart meter and billing system data. Implemented secure and scalable backend functionality including CRUD operations, user registration flows, and structured user profile management. Strengthened application security and session control by adding authentication, role-based authorization, and cache-based session timeout mechanisms to reduce unauthorized access risk.

Research Intern (Remote Data Science Research) at Indian Institute of Technology, Kharagpur

December 1, 2021 - March 1, 2022

Built a scalable NLP dataset (10,000+ B2B software reviews) by building a web crawler and robust preprocessing pipelines for structured and unstructured data. Extracted latent semantic structures and customer segments using hierarchical clustering, K-means, and LDA topic modeling on high-dimensional text embeddings. Modeled multi-level sentiment and topic-importance signals to generate review- and segment-level insights by sentiment analysis and aggregation of topic-weight distributions, contributing to a peer-reviewed paper presented at EMAC 2024.

Remote Data Science Research Intern at Indian Institute of Technology, Kharagpur

December 1, 2021 - March 31, 2022

Engineered a scalable NLP dataset of 10,000+ B2B software reviews by building a web crawler and implementing robust preprocessing pipelines for structured and unstructured data. Extracted latent semantic structure and customer segments to improve interpretability and pattern discovery by applying hierarchical clustering, K-means, and LDA-based topic modeling on high-dimensional text embeddings. Modeled multi-level sentiment and topic importance signals and generated granular review- and segment-level insights via sentiment analysis and aggregation to topic-weight distributions; contributed to a peer-reviewed paper presented at EMAC 2024.