Murad Ali
Ilmenau
Summary
AI Engineer with a strong focus on the applications of Large Language Models and Retrieval Augmented Generation. Experienced in working with Llama, Mixtral, GPT and vector databases to build efficient retrieval systems. Skilled in extracting insights from both structured and unstructured data, including text, documents and audio with experience in preprocessing, cleaning and transformation for AI readiness. Practical experience with tools like LangChain and Haystack and building AI systems that can scale. Comfortable in developing in Agile environments, using Git and cloud based deployments to streamline workflows.
Overview
3
3
years of professional experience
Work History
AI Engineer
Eclevar Medtech
09.2024 - 05.2025
- Worked on AI application development for physicians consultation assistance.
- Leveraged GCP for scalable infrastructure, including Cloud Storage, Cloud Functions and Vertex AI for model deployment and data pipelines.
- Explored and evaluated multiple ASR models including Whisper, Wav2Vec2 and DeepSpeech using Word Error Rate (WER) as the primary metric for accuracy.
- Extracted meaningful insights from transcribed data using both open-source and commercial LLM's including Llama, Mixtral, GPT, Gemini and Claude through APIs.
- Performed data cleaning, preprocessing, post processing and structuring of transcribed text to have high quality data.
- Implemented process automation to improve work intervention.
- Evaluated model responses using tools like RAGAS and Deep Eval to ensure relevance, faithfulness, Precision and Recall.
- Utilized tools like Bitbucket for version control and Jira for project management and task tracking.
Thesis
Fusion Group, Friedrich-Schiller-Universität Jena
03.2024 - 09.2024
- Explored the reproducibility of deep learning methods using large language models.
- Extracted answers to competency questions from text and tables in 100 biomedical scholarly articles using large language models, specifically Mixtral and LLAMA.
- Applied Retrieval-Augmented Generation (RAG) techniques to enhance comprehension and accuracy for complex and long texts.
- Utilized FAISS for efficient similarity search and dense retrieval which enabled relevant context passages in the RAG pipeline.
- Implemented prompt engineering to refine and improve response quality
- Evaluated the RAG responses by measuring Faithfulness, Answer Relevance, Context Precision and Context Recall using Ragas and got exceptionally good scores.
- Applied Agile methodology and Scrum framework for iterative work management.
AI Engineer (NLP/LLMs) Internship
Incowia GmbH
05.2023 - 08.2024
- Used Tesseract OCR to extract text from scanned PDF documents.
- Applied tokenization techniques to break down extracted text into meaningful units.
- Integrated BERT model and transformers into the parsing pipeline to enhance the accuracy and efficiency of address parsing and with BERT's contextual embeddings, the accuracy exceeded 90%.
- Utilized Large Language Models like Falcon and Mistral for extracting data from invoices (tables).
- Maintained version control and facilitated collaborative development using Git.
- Managed project documentation and facilitated team collaboration using Confluence.
Computer Vision Engineer Student Assistant
Friedrich-Schiller-Universität Jena
03.2024 - 09.2024
- Developed a solution for precise object detection and extraction of rulers from images.
- Employed Segment Anything Model (SAM) and GroundingDino for ruler detection, achieving bounding box accuracy above 90%.
- Participated in sprint planning, stand-ups and reviews to maintain project momentum and effectively adapt to changes.
Deep Learning Engineer (HIWI)
Max Planck Institute
04.2022 - 04.2023
- Worked on image classification project with a focus on transfer learning techniques.
- Applied and fine tuned a pre-trained ResNet-50 model for Garlic Mustard image classification improving classification accuracy from 85% to 93%.
- Annotated and labeled a dataset of 10,000 images into growth stages: vegetative, budding, flowering and fruiting.
- Preprocessed and cleaned the dataset using normalization and noise reduction techniques.
Education
Master of Science - Research in Computer and Systems Engineering
Technische Universität Ilmenau
12.2024
Bachelor of Science - Computer Systems Engineering
University of Engineering and Technology Peshawar
07.2018
Skills
- Python
- C
- C
- PyTorch
- TensorFlow
- GPT
- Claude
- DeepSeek
- Gemini
- Llama
- Mistral
- Falcon
- BERT
- ASR
- Scikit-Learn
- Zero Shot Object Detection
- NLTK
- SpaCy
- REST API
- FastAPI
- Hugging Face
- LangChain
- Transformers
- Segment Anything Model (SAM)
- Grounding Dino
- GitLab
- Confluence
Accomplishments
Tech Stack: Python, LangChain, Groq, LLaMA, Jira, REST API, , FastAPI
- Built an autonomous AI agent that interprets natural language requests and automates Jira ticket creation using real-time API calls.
- Integrated LangChain ZeroShotAgent with custom tools for function execution and reasoning.
- Developed and deployed a FastAPI backend to process user requests like "Create a Jira ticket for login error" and return live ticket IDs.
- Implemented conversational behaviour which allows the agent to respond naturally and prompt users for clari
- Handled edge cases, including newline constraints in ticket summaries, API response errors and input ambiguity.
- Created specificaly to explore agentic orchestration, RAG concepts and LLM reasoning.
AI POWERED EARLY DETECTION OF DYSLEXIA IN CHILDREN
Tech Stack: Python, Gemini 1.5 Pro, o3 Mini, ASR, Speech & Handwriting Analysis APIs, Prompt Engineering, Streamlit (optional UI)
- Designed and developed a multi-model application to assist in identifying dyslexia and supporting learning in children, leveraging
- LLMs and VLMs for intelligent interaction and analysis across the following modules.
- Writing: Generated age and language appropriate text using o3 Mini. Implemented image based handwriting analysis using Gemini
- 1.5 Pro to evaluate features like letter reversal, spacing, self-correction and letter formation.
- Phonological Awareness: Enabled users to listen to audio prompts and select matching images, assessing phoneme recognition and
- auditory processing.
- Working Memory: Built tasks requiring users to recall and select image sequences in correct or reverse order based on audio input.
- Reading: Used LLM-generated stories for read-aloud exercises. Employed Gemini 1.5 Pro for speech analysis to detect reading,
- pauses, mispronunciations and hesitation markers.
- Produced an overall probabilistic score indicating the likelihood of dyslexia based on performance across all modules.
Languages
English
German
Timeline
AI Engineer
Eclevar Medtech
09.2024 - 05.2025
Thesis
Fusion Group, Friedrich-Schiller-Universität Jena
03.2024 - 09.2024
Computer Vision Engineer Student Assistant
Friedrich-Schiller-Universität Jena
03.2024 - 09.2024
AI Engineer (NLP/LLMs) Internship
Incowia GmbH
05.2023 - 08.2024
Deep Learning Engineer (HIWI)
Max Planck Institute
04.2022 - 04.2023
Bachelor of Science - Computer Systems Engineering
University of Engineering and Technology Peshawar
Master of Science - Research in Computer and Systems Engineering
Technische Universität Ilmenau
Similar Profiles
Samita SakshiSamita Sakshi
Executive Assistant to MD& CEO at Andhra Pradesh MedTech ZoneExecutive Assistant to MD& CEO at Andhra Pradesh MedTech Zone
SWANUBHAV RATHSWANUBHAV RATH
Product Owner- Web and Digital Platforms at Johnson and Johnson MedTechProduct Owner- Web and Digital Platforms at Johnson and Johnson MedTech
Daniel HillDaniel Hill
Manager, Sports Capital Sales - East at Johnson & Johnson MedTechManager, Sports Capital Sales - East at Johnson & Johnson MedTech
Muskan SharmaMuskan Sharma
Technical Analyst at Kearney India Pvt. LtdTechnical Analyst at Kearney India Pvt. Ltd
Ankur DasAnkur Das
Research Assistant at Technical University of IlmenauResearch Assistant at Technical University of Ilmenau