LLM Researcher
Project Overview
Led research on "Using Large Language Models to Enhance Emergency Response Systems: A Context Awareness Approach". The project aimed to revolutionize 911 systems by integrating AI to support human dispatchers.
Key Responsibilities
- Developed LLM system to analyze 911 call texts and classify emergencies.
- Integrated age recognition from voice inputs to tailor LLM recommendations.
- Drafted comprehensive operational guidelines manual ensuring alignment with best practices.
Achievements
- Presented research at the National Conference on Undergraduate Research (NCUR) 2025 in Pittsburgh, USA.
- Engaged with over 60 attendees during a dedicated two-hour poster session.
- Achieved 76.5% accuracy in Address Extraction and 48.3% in Name Recognition.
Technologies Used
- LLMs: LLaMA 2, GPT-4
- NLP: BERT-large-cased for NER
- Speech-to-Text: OpenAI Whisper
- Frameworks: Python, PyTorch, Hugging Face
System Architecture
| Component | Technology | Purpose |
|---|---|---|
| Speech-to-Text | OpenAI Whisper | Real-time call transcription |
| NER | BERT-large-cased | Extract caller info & location |
| Classification | Fine-tuned LLM | Determine emergency type |
| Response Gen | Context-Aware LLM | Generate dispatcher protocols |