AAI_2025_Capstone_Chronicles_Combined

‭ResolveAI‬

‭Bidirectional LSTMs offer significant advantages for ticket priority classification by‬ ‭enabling the model to capture context from both preceding and succeeding words in a text‬ ‭sequence, which is critical in understanding the full semantic meaning of a ticket’s content‬ ‭(Anishnama, 2023). In many ticketing scenarios, vital cues indicating urgency or severity may be‬ ‭scattered throughout the text i.e. key phrases at the beginning might set a context that is clarified‬ ‭or even contradicted later, and critical urgency terms could appear at the end of the description.‬ ‭By processing the input in both forward and backward directions, a bidirectional LSTM‬ ‭effectively combines these perspectives, yielding a richer and more nuanced representation of the‬ ‭language. This dual-context approach helps the model differentiate between similar phrases that‬ ‭imply different priorities, deal with ambiguous or implicit cues, and better handle‬ ‭domain-specific terminology. Moreover, by capturing long-range dependencies and subtle shifts‬ ‭in tone or emphasis across the entire text, bidirectional LSTMs contribute to a more robust and‬ ‭reliable classification performance, ultimately leading to improved decision-making in the‬ ‭assignment of ticket priorities.‬ ‭Another key component of the overall pipeline is the retrieval augmented generation‬ ‭(RAG) enhanced response system, where we leverage LangChain and ChromaDB to enhance‬ ‭LLM-generated responses for low priority tickets. LangChain serves as the core framework for‬ ‭orchestrating the LLM’s interactions, retrieving relevant information from ChromaDB, prompt‬ ‭construction and response generation (Langchain,2025). ChromaDB is responsible for storing‬ ‭data as vectors, enabling speedy semantic searches that help retrieve the most relevant content‬ ‭for a given query. There are a plethora of different types of queries that users may pose, ranging‬ ‭Additional Components‬

‭12‬

60