Falcon Market — Student Exchange Platform
Marketplace for NYUAD students to exchange campus dining currency, compare rates, and connect through verified profiles.
I research secure and efficient AI systems, and build dependable software that turns ambitious ideas into useful products.

Focus
AI Security · ML Systems
I'm Hailemariam — a Computer Science student and Applied Mathematics minor at NYU Abu Dhabi. I study how language models behave under adversarial pressure and how efficiently their generated programs run.
At SANAD Lab, I researched prompt injection in autonomous blockchain trading agents and built PerfArena, a reproducible benchmark for the correctness, runtime, memory, and energy use of LLM-generated code. At eBrain Lab, I co-authored an EMNLP 2026 submission on multimodal jailbreak attacks.
Outside research I ship full-stack products, including Falcon Market, a secure student exchange platform, and Locify, an AI-native tour guide. I also serve as a Residential Assistant and Spiritual Life Ambassador on campus.
B.S. Computer Science · Minor in Applied Mathematics
NYU Abu Dhabi · Aug 2023 – May 2027
Abu Dhabi, UAE
Relevant coursework
GitHub Activity
Right now
LLM security and generated-code efficiency
SANAD Lab & eBrain Lab · 2026
5+
research projects
7+
shipped projects
3+
years coding
Selected work spanning full-stack systems, applied AI, and computer vision.
FALCON MARKET
NYUAD student exchange
BUY
2.14
current rate
SELL
2.08
current rate
01 / 03 · 2026
Problem
Built and deployed a student marketplace for exchanging campus dining currency, comparing current rates, and connecting through public profiles. Implemented NYU-only Google authentication and PostgreSQL row-level security to enforce account eligibility and data ownership.
Approach
Added offer editing and renewal with seven-day expiry, hourly rate-history snapshots, persistent dark mode, and database and browser tests for core workflows..
Result
See the code.
● AI narrating nearby POIs…
Sheikh Zayed Grand Mosque — 0.4 km
02 / 03 · 2025
Problem
Built at HackPrinceton Fall 2025, maptourai is a real-time AI tour guide. It discovers nearby historical POIs via Foursquare + Brave search, then streams context-aware narrations through GPT-4.1/Claude via MCP agents.
Approach
The backend is a FastAPI service with location-aware caching; the frontend is Next.js with Mapbox GL for smooth map interactions..
Result
See the code.
precision
0.60
recall
0.73
03 / 03 · 2025
Problem
Fine-tuned the DETR (Detection Transformer) model on pixel-difference frame pairs to detect moved objects in video. Ran systematic ablations on learning rate, backbone freezing, and augmentation strategies.
Approach
Best checkpoint achieved val precision 0.60 and recall 0.73, with automated evaluation visualizations for every run..
Result
See the code.
Undergraduate research spanning LLM security, code efficiency, climate science, and NLP.
Both demos run entirely in your browser — no server, no data sent anywhere.
Runs locally in your browser — no server
Try examples:
Benign:
Malicious:
Semantic drift across transformer layers
Select a polysemous word:
Sentence A (financial sense):
I need to go to the bank to deposit my paycheck.
Sentence B (different sense):
We went fishing by the river bank at sunset.
8 projects across ML, systems, and the web.
Marketplace for NYUAD students to exchange campus dining currency, compare rates, and connect through verified profiles.
AI-native tour guide that tracks user location, caches POIs, and serves narrations generated via Dedalus MCP agents.
Built a mini-CNN framework with numeric gradient checks; fine-tuned ResNet18 on EMNIST, reaching 96.5% test accuracy.
Fine-tuned facebook/detr-resnet-50 on pixel-diff frame pairs; best val precision 0.60, recall 0.73.
Typing game with auth, difficulty levels, and live leaderboards tracking WPM and accuracy; deployed on Render.
Full-stack platform with AI essay feedback, Common App guides, and peer-review workflow; 200+ students in first month.
RESTful task manager with user auth, priority/status filters, and containerized deployment.
Hungarian-matching pipeline combining IoU, centroid distance, and class cues for multi-object tracking.
Research, engineering, and leadership across NYU Abu Dhabi and beyond.
Open to research collaborations, internships, and interesting conversations about ML security, computer vision, or anything in between.
Email me →Or press ⌘K to open the terminal and try sudo hire-me.