Research project for task-oriented dialogue system with jointly training multi-intent classification and slot filling

MSc project: design, develop and evaluate a conversational story-teller dialogue system. The system is developed in python and makes use of a Telegram web interface. The connection between the system and the web interface is done by using a SSL connection provided by Ngrok. The system includes a BERT model with a span classification head on top to handle user questions. A naive bayes classifier is used for the sentiment analysis task and a Rasa intent classifier is used to recognise user intents.

Agentic DSPy RAG is a production-ready, multi-agent Retrieval-Augmented Generation (RAG) system built with the DSPy framework. It features vision-based data ingestion, advanced intent classification, specialized agent routing, multi-step reasoning, and a robust FastAPI backend. Designed for high-quality, context-aware conversational AI over complex

Customer Service Voice Response System with 3-layer intent classification pipeline combining rule-based NLP, neural semantic embeddings (SBERT), and LLM. (Speech Recognition and Synthesis)

In a task oriented domain, recognizing the intention of a speaker is important so that the conversation can proceed in the correct direction. This is possible only if there is a way to label the utterance with its proper intent. One such labeling technique is Dialog Act (DA) tagging. The main goal of this thesis is to build a Dialog Act tagger for the Telugu English Code Mixed corpus. Dialogue Act (DA) classification plays a key role in dialogue interpretation, especially in spontaneous conversation analysis. Dialogue acts are defined as the meaning of each utterance at the illocutionary force level. Code-Mixing (CM) is a very commonly observed mode of communication in a multilingual configuration. The trends of using this newly emerging language have its effect as a culling option especially in platforms like social media. This becomes particularly important in the context of technology and health, where expressing the upcoming advancements is difficult in native language. Despite the change of such language dynamics, current dialog systems cannot handle a switch between languages across sentences and mixing within a sentence. Everyday conversations are fabricated in this mixed language and analyzing dialog acts in this language is very essential in further advancements of making interaction with personal assistants more natural. Almost all standard traditional supervised machine learning approaches to classification have been applied in DA classification, from Support Vector Machines (SVM), Naïve Bayes, NLTK Classifiers, Max Entropy Classifier, Multilayer Perceptron, Conditional Random Field Classifier and Hidden Markov Model (HMM).

Benchmarking tool for various intent and entity classification systems

Question Embeddings Based on Shannon Entropy. Solving intent classification task in goal-oriented dialogue system

Intent of this project is the rapid development of a neural network for image classification. Thanks to the use of framework like Keras this is possible by moving the first steps using refinement techniques starting from known models. There is discussion of the architecture of a USB commercial device, Intel Movidius neural compute stick, with low power consumption for neural network execution on SoC systems such as Raspberry. Finally, there are the problems and limitations that occurred during the development and distribution of the software implemented.

intent-classification for Dialogue system

Intent-routed, local-first multimodal assistant that combines ONNX Flan-T5 Large, MiniLM intent classification, hybrid RAG (ChromaDB + BM25), web search gating, YOLO-NAS camera tools, voice mode and a modern web UI into a single production-style system.

A baseline system for intent recognition using the dataset released by SMP2018 ECDT Shared Task

Implemented an NLP-based intent classification system using text preprocessing, vectorization, and supervised learning to map user queries to predefined intent categories.

NLP-based chatbot using NLTK and Keras with intent classification, confidence filtering, suggestion system, and chat logging.

Production-ready MLOps system for intent classification with FastAPI, Apache Airflow, Docker, CI/CD, and automated drift detection.

Built an end to end dialogue system for intent classification and slot filling to find restaurants and hotels.

Multilingual Intent & Threat Classification system using Hugging Face Transformers to detect hate, anti-state propaganda, panic spreading, misinformation, and normal expressions.

This project is an Intent Classification System built using Machine Learning (Naive Bayes) and TF-IDF Vectorization. It classifies user inputs into predefined categories (intents) such as greetings, farewells, or requests.

CrewAI multi-agent router system with local LLM integration (Ollama + Llama 3.1). Intelligent intent classification with specialized file and math agents.

Built end-to-end AI customer service system achieving 96% intent classification accuracy, featuring multi-intent detection, entity extraction, and automated priority handling - reducing response time from hours to milliseconds while maintaining professional, context-aware responses.

An end-to-end PEFT-based intent classification system that reduces RAG inference costs by 95% and latency by 70% compared to GPT-4 routing.

Conversational AI system for voice-driven accounting that converts Hindi/Hinglish speech commands into structured financial actions using an NLP intent classification and entity extraction pipeline.

A machine learning phishing detection system trained on structured email and URL features. MailHarpoon leverages feature engineering and supervised classification to identify malicious intent in digital communication.

This dataset contains a curated collection of **phishing and legitimate (benign) emails** for use in cybersecurity training, phishing detection models, and email classification systems. Each entry is structured with subject, body, intent, technique, target, and classification label.

An intent classification system (7 categories) using both TF-IDF + Logistic Regression and a fine-tuned DistilBERT model on the SNIPS dataset. The DistilBERT version is on Streamlit for demoing!

A robust multilingual text intent classification system that predicts user intent across five categories — Hate/Violence, Anti-state Propaganda, Panic Spreading, Misinformation, Neutral. Built with Torch and Transformers, it supports 20+ languages, works fully offline, and is ideal for secure, low-connectivity, or production environments.

An AI chatbot demonstrating NLP, intent recognition, and dynamic learning. It uses fuzzy matching for input detection, JSON-based runtime knowledge updates, and real-time conversational learning. This project lays the foundation for building advanced systems like ChatGPT by exploring key concepts such as intent classification, NLU, and adaptive AI.

Natural language understanding courses assignments 📌 Exe1: Persian Poem Meter Classification 📌 Exe 2: Intent Detection & Slot Filling 📌 Exe 3: QA System with RAG 📌 Final Proj: Multi-Turn Chatbot for Banking 🏦💬

This is an an autonomus booking AI agent system that work with hybrid LLM architecture of local Ollama deepseek chat model ,open ai api for intent classification and emmbedding ,and gemeni for refinement in response .

Smart customer assistant demo showcasing multimodal AI: text/image/speech processing with intent classification & automated responses. Built with PyTorch, Transformers, OpenCV. Trained on synthetic data for demonstration; full production system deployed at scale in live environments.

A sophisticated Natural Language Processing (NLP) system specifically designed for medical text analysis. This pipeline combines state-of-the-art NLP models to extract meaningful information from medical texts, including patient information, conditions, temporal data, and intent classification.