L
Lokesh Madem
About
Detail
Texas, United States
Résumé
Jobs
verified_user
0% verified
ML EngineerTubiMay 2025 - Current (1 year 6 months)• Prototyped an LLM summarization pipeline using prompt engineering and schema-constrained decoding to generate spoiler-safe title summaries; stored outputs and embeddings in Weaviate to support retrieval for mood-based recommendation rows. • Curated and preprocessed content inputs (trailers, synopses, reviews) with light normalization and safety filtering to support consistent, schema-valid generation at scale. • Partnered with retrieval and ranking teams to integrate summaries into offline experiments, improving long-tail recall and enabling cold-start support for homepage recommendation rows.- AI EngineerPayPalJun 2024 - Apr 2025 (11 months)• Developed a Retrieval Augmented Generation (RAG) chatbot using L L M s (GPT, Llama, and Mistral) models with Pinecone for vector storage, enabling context-aware, real-time customer support. • Built the retrieval engine, including multi-format chunking, embedding generation (all-MiniLM), and semantic search optimization for fast and accurate top-K retrieval. • Fine-tuned Llama models using LoRA adapters and supervised fine tuning on curated datasets to align with PayPal's domain-specific language and compliance standards, yielding ~10% higher human-judged answer correctness and ~15% lower hallucination rate on an internal 500-query eval set. • Orchestrated model deployment by integrating AWS Lambda, SageMaker, and ECS/StepFunctions for
- LAI EngineerL & T technology servicesApr 2021 - Jul 2022 (1 year 4 months)• Developed object detection models, including Faster R-CNN and YOLO using PyTorch, to detect and recognize worker safety equipment like helmets, reducing safety incidents by over 25% and increasing workplace compliance monitoring speed by 40%. • Managed the collection and annotation of over 10,000 images of workers in industrial environments, collaborating with domain experts to train and validate AI models. • Implemented data preprocessing and augmentation techniques that resulted in a 20% improvement in model robustness and performance. • Set up automated testing frameworks to continuously track the system's mAP and other relevant metrics such as precision, recall, and F1-score, enabling quick identification of issues during developm
Education
verified_user
0% verified
- Master of Science;University of North TexasAug 2022 - May 2024 (1 year 10 months)
- Bachelor of Technology;Andhra UniversityJul 2018 - May 2022 (3 years 11 months)
- Neural Network & Deep LearningCoursera
- Python for Everybody SpecializationCoursera
- Improving Deep Neural Network: Hyperparameter Tuning, Regularization & OptimizationCoursera
Projects (professional or personal)
verified_user
0% verified
- UUniversal UI Workflow AgentNov 2025 - Dec 2025 (2 months)• Built an LLM-powered browser agent using Python, Playwright, and OpenAI APIs to translate natural
language instructions into executable web actions, supporting workflow automation across Notion, Jira,
Asana, and Trello.
• Implemented self-healing automation with DOM parsing, accessibility-tree metadata, execution tracing,
screenshot lo - AAutomatic Image CaptionerAug 2023 - Dec 2023 (5 months)Designed and implemented an automatic image captioning system using natural language processing and computer vision techniques to generate descriptive captions for images, demonstrating strong proficiency in machine learning and image processing.
- HHeart Disease Prediction System
- AAutomatic Gym Trainer
- LLanguage TranslatorTechnology Used: NLP, AWS, Transformer, Docker • Developed a real-time language translation model using transformer architecture in PyTorch, trained on a dataset of 0.45 million English- Telugu sentence pairs, achieving a BLEU score of 0.55. • Deployed an end-to-end MLaaS framework for English-to-Telugu translation, encompassing data acquisition, preparation, model tuning, and API implementation with Flask and Postman. • Successfully hosted the model on the Waisum platform (https://www.waisum.co/), a collaborative MLaaS environment supporting blog hosting and ML model prototyping with a capacity for 1000+ concurrent users. Technology used: Com