01
🥇 1st Place · BSEAS FYP Open House 2026
VisiHealth AI
Medical Visual Question Answering System · Final Year Project
Built a full MedVQA system using a Dual-Encoder + Multi-token Cross-Attention architecture
with BioLinkBERT, trained on the SLAKE dataset. Also engineered the backend APIs for model inference, ROI
localization, and data processing. This one won first place at the CS department showcase.
PythonFastAPIBioLinkBERTSLAKEMedVQA
class MedVQAModel:
def __init__(self):
self.encoder = BioLinkBERT()
self.attention = CrossAttn()
def answer(self, img, q):
v = self.encode_image(img)
t = self.encode_text(q)
return self.fuse(v, t)
VisiHealth AI · Python
02
AccessHire
AI-powered job portal for visually impaired users
Built the entire backend — REST APIs, JWT auth, user & job management, and a content-based recommender system. Integrated AI speech-to-text & text-to-speech via a FastAPI microservice for a fully accessible experience.
NestJSFastAPIMongoDBJWTSpeech AIRecommender
03
VelocityMD
Doctor appointment scheduling backend
Designed the full scheduling logic and RESTful APIs — appointments, conflict
detection, and user management. Nothing fancy on the surface, bulletproof under the hood.
FastAPIRESTSchedulingAuth
04
Metro Ticketing
MERN stack ticketing platform
Full-stack metro booking system — from seat selection to ticket confirmation. My
first real taste of full-stack. Mongo, Express, React, Node. Shipped it.
MongoDBExpressReactNode.js
05
Narrative Mutation Tracker
RAG-powered misinformation pipeline
Traces how viral news claims mutate across media sources — origin detection, narrative drift scoring, and source credibility analysis. Built on a full RAG pipeline with semantic search and LLM-powered reasoning.
RAGLangChainFAISSPythonFastAPI
