just a human trying to understand structure in a universe full of noise.

I study Mathematics & Computer Science and spend most of my time trying to understand how intelligent systems see, reason, remember, and fail.

My work lives somewhere around Vision-Language Models, Topological Data Analysis, Graphs, and reliable AI — especially when the problem has structure that a neural network would rather ignore.

I like building things that are not just bigger, but a little more thoughtful about how information is represented and connected.

medical vision-language models
topological & structural reasoning
graph-based intelligence
agentic retrieval systems
reliable / less-hallucinatory AI
3D, spatial & multimodal reasoning

A recurring question behind most of my work:

Can we give neural models better structural priors instead of expecting scale to discover everything?

Medical VQA · Vision-Language Models · Topological Data Analysis

Teaching a generative medical VLM to reason about shape, spatial structure, and topology in gastrointestinal endoscopy images.

Patch-level TDA features are fused into both the visual representation and LLM decoder through gated adapters.

BLEU 0.477 · ROUGE-L 0.691 · METEOR 0.695

→ medico_vqa_2026

Neo4j · Qdrant · LLM Agents · Graph Reasoning

A multi-agent academic tutor that navigates knowledge through concept graphs instead of blindly retrieving chunks.

It uses prerequisite-aware Reverse BFS, hybrid memory, HyDE retrieval, and specialized Math / Algorithm / Concept / Example agents.

RAGAS 0.902 · Faithfulness 0.964

→ rag-expert-mentor

Program Analysis · Runtime Graphs · LLM Test Generation

An AI-powered PR test coverage system that maps code changes to runtime behavior and generates tests with execution-level evidence.

Because "the test looks reasonable" is not quite the same as "the test actually covers the function."

6/6 generated tests passed · 336 runtime edges · 0 missing coverage

→ softgnn-advisor

📄 CATA: Clinical-Aware Topological Adaptation for Generative Medical VQA MediaEval 2026 · co-located with ACM ICMR 2026

Top 2 — Explanation Task Top 4 — Answer Task

→ paper

languages = ["Python", "C", "C++", "Java", "R", "MATLAB"]

ml = [
    "PyTorch",
    "TensorFlow",
    "scikit-learn",
    "NumPy",
    "Pandas"
]

systems = [
    "Linux",
    "Docker",
    "Git",
    "Neo4j",
    "Qdrant"
]

currently_obsessed_with = [
    "Vision-Language Models",
    "Topological Data Analysis",
    "Graph Neural Networks",
    "Agentic AI",
    "Structural Reasoning"
]

When I'm not staring at tensors, graphs, or papers, I'm probably somewhere around:

🌌 space · 🏎️ Formula 1 · 🤖 robots · 🎮 sci-fi games 🎸 guitar · 🎹 piano · 🎬 action / science fiction

I have an unreasonable appreciation for machines that move fast, fictional universes that are unnecessarily large, and mathematical ideas that somehow end up inside neural networks.

I'm always interested in conversations around AI research, multimodal reasoning, topology, graphs, and strange ideas that might actually work.

LinkedIn · Email