K
Koushik Sai Guttula
About
Detail
Seattle, Washington, United States
Résumé
Jobs
verified_user
0% verified
- UAI EngineerUltimus SolutionsJun 2025 - Apr 2026 (11 months)• Engineered a Retrieval-Augmented Generation (RAG) pipeline utilizing FAISS/Chroma to retrieve relevant FAQs, product documentation, and past resolutions, accelerating response efficiency by 30%. • Fine-tuned and assessed ML/NLP classifiers (Logistic Regression, Random Forest, XGBoost, and Transformers), achieving 88% ticket classification accuracy and 91% urgent-ticket recall. • Streamlined priority scoring and ticket routing through semantic similarity, urgency detection, and risk keyword analysis, reducing manual triage effort by 25%. • Architected prompt-engineering workflows for GPT-based systems, incorporating intent detection, entity extraction, and sentiment analysis while optimizing prompts to reduce token consumption. • Establish
AI EngineerIntone NetworksJan 2024 - Jul 2024 (7 months)• Deployed Retrieval-Augmented Generation (RAG) solutions leveraging GPT-based models, vector databases (FAISS, Pinecone), and semantic search techniques to support intelligent knowledge retrieval across large document repositories. • Formulated prompt-engineering and prompt-optimization frameworks, improving response relevance by 35% through systematic assessment of retrieval quality, hallucination rates, and answer-grounding metrics. • Constructed transformer-based NLP workflows utilizing BERT embeddings and sentence transformers for semantic search, intent classification, and contextual question answering. • Executed experimentation and statistical analysis on user interaction data, validating model improvements through hypothesis testin- DAI EngineerDatahub TechnologiesJun 2023 - Dec 2023 (7 months)• Crafted deep learning models for text classification and spam detection using TensorFlow, PyTorch, LSTM, CNN, and Transformer architectures. • Orchestrated end-to-end machine learning workflows encompassing feature engineering, TF-IDF vectorization, embedding generation, model training, evaluation, and deployment. • Refined data preprocessing workflows through parallel processing and efficient feature extraction techniques, reducing model training time by 40%. • Investigated comparative performance across advanced machine learning models and deep neural networks, leveraging statistical significance testing and model evaluation metrics to identify optimal architectures. • Delivered real-time inference services using FastAPI and Docker, ena
Education
verified_user
0% verified
- UMaster of Science in Artificial IntelligenceUniversity of Michigan - DearbornAug 2024 - Apr 2026 (1 year 9 months)CGPA: 3.94
- Bachelor’s in Computer Science and EngineeringMahindra UniversitySep 2020 - May 2024 (3 years 9 months)CGPA: 3.4
Projects (professional or personal)
verified_user
0% verified
- Language Detection & Translation AppGitHubJan 2024Tools: Python, TensorFlow, Streamlit, Google Translate API, gTTS, NLP • Engineered a TensorFlow-based multilingual language classification system leveraging tokenization, sequence padding, embedding layers, and dense neural networks to automatically identify input text across multiple supported languages. • Integrated a full-stack NLP workflow using Streamlit, combining language detection, real-time translation through Google Translate APIs, and confidence-scored inference to deliver an interactive multilingual experience. • Enabled speech synthesis functionality using Google Text-to-Speech (gTTS), generating audio output for translated content and delivering a seamless text-to-translation-to-speech pipeline.
- News Article SummarizationGitHubJan 2024Tools: Python, PyTorch, Hugging Face Transformers, T5, spaCy, ROUGE, Weights & Biases, Tkinter, CNN/DailyMail Dataset • Fine-tuned a T5-based abstractive text summarization model using Hugging Face Transformers on the CNN/DailyMail dataset, processing 287K+ training articles with tokenization, truncation, padding, and sequence-to-sequence learning workflows. • Established reproducible training and evaluation pipelines using PyTorch, Hugging Face Trainer, GPU acceleration, and Weights & Biases experiment tracking, achieving validation loss improvement from 0.2661 to 0.2640 across three training epochs. • Created an end-to-end summarization application supporting both extractive and abstractive approaches using spaCy, T5, Tkinter, and ROUGE-b
- Hybrid Semantic Search EngineGitHubJan 2024Tools: Python, BM25, TF-IDF, NetworkX, PageRank, Cosine Similarity, Pandas • Architected a hybrid information retrieval engine on 20K Wikipedia articles using BM25 candidate retrieval, TF-IDF vectorization, cosine similarity, PageRank authority scoring, and title-based relevance boosting to enhance search ranking quality. • Formulated a two-stage retrieval and re-ranking framework that identifies high-relevance candidates through probabilistic BM25 scoring and refines rankings using lexical relevance, vector-space similarity, graph authority, and query-title matching signals. • Devised a document similarity graph using TF-IDF cosine similarity and NetworkX PageRank to incorporate global authority signals, while generating query-aware snippe