Arman A Anwar

This is a proof-of-concept system that generates unmanned aerial vehicles (UAV) flight plans from natural language input. It requires advanced reasoning to produce legal, safe, and Part 107-compliant missions. Unlike typical chat-based automation, this system handles complex mission planning with regulatory, geographic, and operational constraints.

I architected a hybrid multi-agent system using OpenAI and LangChain to extract structured mission parameters (e.g., location, altitude, and type) via intent parsing and slot-filling. I also integrated OpenStreetMap (OSM) queries to inform planning decisions with real-world terrain, no-fly zones, and structural data. I used retrieval-augmented generation (RAG) to incorporate relevant prior missions into flight plan generation, improving planning efficiency and contextual relevance. In addition, I prototyped a constraint-solving agent using SciPy to optimize UAV allocation and enable parallel mission execution, implementing validation routines to simulate and check generated flight plans for LLM hallucination mitigation and operational safety.

A DARPA-funded project to explore early detection of PTSD and traumatic brain injury (TBI) in veterans through analysis of social media content. The system was designed to learn directly from training inputs provided by medical professionals, enabling adaptive screening and potential early intervention strategies.

I developed a machine learning pipeline capable of extracting and analyzing linguistic patterns indicative of PTSD and TBI from unstructured social media text. The pipeline incorporates expert-labeled training data from clinicians to guide supervised model development and evaluation. I addressed domain-specific challenges such as limited labeled data, informal language, and privacy-preserving data handling. Springer recognized the project outcomes in "Sentiment Analysis for PTSD Signals," highlighting the system's contribution to computational mental health research.