Suneetha Cherukuwada

Machine Learning Engineer (AI & GenAI solutions)

About

Machine Learning Engineer with over 4+ years of experience designing and deploying scalable AI solutions across both natural language processing (NLP) and computer vision domains. My work combines deep technical skill with a practical mindset, building real-world systems that improve performance, reduce latency, and deliver measurable business impact. In the NLP space, I’ve developed and fine-tuned transformer-based models including GPT-2, LLaMA-2, and FLAN-T5, using techniques like LoRA, quantisation, and prompt engineering. My projects include a medical document summarisation API that achieved over 85% BERTScore, a domain-specific chatbot for clinical texts, and a legal contract analysis tool powered by retrieval-augmented generation (RAG). These solutions integrated semantic search, vector similarity, and custom datasets to ensure accuracy, reduce hallucinations, and provide actionable insights in real time. My computer vision work has addressed complex industrial and automation challenges. I’ve built systems for vehicle detection using U-Net, automated damage detection using GoogLeNet and YOLO, and screw-fastening verification for automotive assembly using quantised ConvNets on Raspberry Pi. I also fine-tuned YOLOv10 to detect and count signage on construction drawings, streamlining bill-of-quantity estimation processes. These models were deployed with ONNX, Docker, and FastAPI, often optimised for edge devices and cloud environments such as AWS ECS and SageMaker. I specialise in building end-to-end ML pipelines, from data preprocessing and model training to deployment, monitoring, and CI/CD integration. My toolkit includes Python, PyTorch, Hugging Face Transformers, TensorFlow, MLflow, and cloud platforms like AWS and GCP. I’m also experienced in MLOps practices, enabling reproducible experiments and scalable deployment workflows through Docker and GitHub Actions. What drives me is the challenge of turning unstructured data into intelligent systems that work reliably in production. Whether it’s improving inference latency on constrained hardware or enabling semantic understanding of complex documents, I focus on designing systems that are robust, efficient, and impactful. I’m always open to connecting with professionals, researchers, or organizations exploring cutting-edge applications in AI. If you’re working on something exciting at the intersection of NLP, computer vision, and MLOps, I’d love to hear from you.

Country

Germany

City

Munich

Industry

Computer Software

Skill

Machine Learning Algorithms, Large Language Model Operations (LLMOps), Continuous Integration and Continuous Delivery (CI/CD), Data Modeling, Modeling, Computer Science, Prompt Engineering, Generative AI, Transformer Models, Problem Solving, AI Software Development, Hugging Face Products, k-means clustering, Motion Tracking, Object Detection, MLflow, Natural Language Processing (NLP), Deep Neural Networks (DNN), Transformers, Large Language Models (LLM)

Experience

Machine Learning Engineer (AI & GenAI solutions)

2023-3 - Present · 3 yrs 7 mos

- Parking Lot Segmentation: Developed a real-time Vehicle segmentation system using U-Net to process CCTV feeds, detecting vacant & occupied spaces. Integrated morphological operations for accuracy & deployed as ONNX model with FastAPI endpoint for seamless real-time inference. - Signage detection & Counting: fine tuned YOLO-v10 for detecting a set of signs on floor plan layout images, enhancing the accuracy in estimating bill-of-quantities. deployed the model as a scalable AWS ECS task, enabling efficient inference with automated load balancing and autoscaling. - On-device deployment: optimized heavy memory footprint YOLO model using static model quantization to deploy onto raspberry-pi. - Developed a real-time summarisation API for clinical notes, discharge summaries and patient reports using a fine-tuned LLaMA-2 model with LoRA. Trained on healthcare data, optimized with 8-bit quantisation, and deployed via AWS SageMaker and FastAPI. Ensured medical relevance with ROUGE and BERTScore evaluation. - Built a RAG-based QA system using Hugging Face Transformers to assist legal professionals in querying complex contract documents with accurate, context-aware responses. Leveraged the CUAD dataset, SentenceTransformers for semantic retrieval, and FLAN-T5 for generative answering, with GPU acceleration to ensure responsiveness and scalability. Applied advanced NLP techniques in legal text understanding, vector similarity search, and transformer-based QA pipelines. - Developed a chatbot that enables chatting with medical documents, achieved by fine-tuning a GPT‑2 model with Hugging Face Transformers and PyTorch. Crafted a custom dataset with expert responses and applied advanced tokenisation and training techniques using the Trainer API. Deployed a full-stack solution with Flask.

Machine Learning Engineer

NEXUSTEC GmbH

2018-10 - 2022-9 · 4 yrs

Munich, Bavaria, Germany

•Engineered a real-time computer vision system to detect screw fastening accuracy using advanced video tracking techniques, enhancing manufacturing precision. • Deployed Siamese-based ConvNets (CNNs) for detecting damaged car parts in the process of assembling, improving defect identification & reducing manual inspection time. • Hosted an auto-annotation tool based on CVAT, significantly optimising in-house video labeling & reducing manual annotation efforts. • Explored cutting-edge deep learning architectures, including YOLO, Siamese Networks, ResNet, & GoogLeNet, to tackle complex computer vision challenges in industrial automation. • Deployed ML models on edge devices, such as Raspberry Pi, optimizing real-time inference in resource-constrained environments.

Project Intern

ISRO - Indian Space Research Organization

2018-1 - 2018-3 · 3 mos

Secundrabad

• Implemented ConvNet based system for Ships detection from Satellite imagery. • Curated data from public & academic sources & trained the system to improve the performance on production data