I am Md. Hamid Hosen, a Software Engineer (Mobile) at
Project 2morrow Software Ltd. (P2M) and a Lead Researcher at
ELITE Research Lab LLC.

I develop scalable and high-performance mobile applications using Flutter, Dart, Kotlin, Jetpack Compose, and modern mobile software architectures.

Alongside software engineering, I conduct research in Large Language Models (LLMs), Mathematical Reasoning, Computer Vision, Multimodal AI, Explainable AI, and Natural Language Processing.

My work focuses on connecting advanced AI research with reliable, production-ready software systems.

• 💼 Software Engineer (Mobile) at P2M Software Ltd.
• 🔬 Lead Researcher at ELITE Research Lab LLC
• 🎓 B.Sc. in Computer Science and Engineering
• 🏫 East Delta University, Bangladesh
• 🌍 Interested in international M.Sc. and Ph.D. research opportunities
• 🤝 Open to AI research, software engineering, and open-source collaboration

Recognized as a Prize Winner in the AI Mathematical Olympiad Proof Pilot, an invitation-only AI research evaluation hosted on Kaggle as part of the AIMO Prize initiative.

• One of only seven invited teams worldwide
• Worked with fully open-source Large Language Models
• Developed mathematical reasoning and proof-generation workflows
• Conducted inference optimization and reproducible evaluation
• Prepared containerized deployment and technical reports

Earned a Gold Medal in the AI Mathematical Olympiad Progress Prize 3.

• Global Rank: 14th
• Overall Score: 43.5
• Competed among 4,138 teams worldwide

Received the Hardest Problem Prize for being the only contestant among 4,138 teams to solve the highly resistant mathematical problem internally referred to as “ACUTES” in both evaluation attempts.

40+ competitions entered across mathematical reasoning, program synthesis, medical imaging, remote sensing, molecular identification and fraud detection.

Detecting coordinated player pairs (chip dumping, soft play, coordinated isolation) in 2,000,000 synthetic No-Limit Hold'em hands — 12,000 players, 18.6 M ordered actions, 112,540 evaluation pairs — and retrieving the specific hands that prove it. Metric: 0.7 · pair AP + 0.2 · evidence MAP@5 + 0.1 · behaviour MAP.

• Team Hack2Publish — designed and built the full pipeline (18 iterations), from 0.67 to 0.900 on the public leaderboard (public baselines plateau at 0.69)
• Behaviour-clone policy model: a 6-class LightGBM trained on all 18.6 M actions (cards, exact 7-card strength, board texture, position, pot/stack ratios, each player's other-phase style) turns every decision into a policy-surprise score — coordination appears as two partners acting wrongly for their cards in the same hand
• Positive–unlabelled pair ranking with exposure-matched training windows, other-phase pair baselines, coincidence z-scores and hard-negative retention; final risk from 20-seed full-data bags
• Two-stage evidence retrieval (pair-hand features → pooled re-ranker with made-hand combinations, street timing, per-player policy surprise and action-sequence templates)
• Fully reproducible: one-command rebuild, ≤1,500-word write-up, five investigator-style case reviews, MIT-licensed code

ARC Prize 2026 Paper Track · NVIDIA Nemotron Model Reasoning Challenge · Gemma 4 Good Hackathon · Google Code Golf 2025 · NeurIPS Open Polymer Prediction 2025 · Biohub Cell Tracking During Development · Neurogolf 2026 · CSIRO Biomass · ROGII Wellbore Geology Prediction · Stanford RNA 3D Folding · Stanford Ribonanza RNA Folding · CZII Cryo-ET Object Identification · ISIC 2024 Skin Cancer Detection · RSNA 2024 Lumbar Spine · RSNA 2023 Abdominal Trauma · BirdCLEF 2025 · LMSYS Chatbot Arena · LLM — Detect AI-Generated Text · Kaggle LLM Science Exam · Eedi Mining Misconceptions in Mathematics · MAP Charting Student Math Misunderstandings · CommonLit Evaluate Student Summaries · Bengali.AI Speech · Home Credit Credit Risk Model Stability · Optiver Realized Volatility Prediction · Child Mind Institute Problematic Internet Use · Drawing with LLMs · Pokémon TCG AI Battle Challenge · AI Agent Security — Multi-Step Tool Attacks · March Machine Learning Mania 2025 / 2026 · Hyperspectral Object Detection & Tracking 2026 · UMUD Muscle Ultrasound Challenge · CUHK-X Small / Large Model Tracks · Soil Grain Size from Photos · Filament Segmentation 2026 · Playground Series S6E7

Ranks marked in progress are current public-leaderboard positions (September 2026) and will be updated when those competitions close.

• 🧠 Large Language Models
• ➗ Mathematical Reasoning
• 🔀 Multimodal AI and Multimodal Fusion
• 👁️ Computer Vision
• 🔍 Explainable and Trustworthy AI
• 💬 Natural Language Processing
• 📱 AI-powered Mobile Applications
• 🩺 Medical AI and Mobile Health
• 🎨 Vision-based UI Understanding
• ⚙️ Reproducible AI Evaluation
• 🚀 AI Inference and Optimization

Project 2morrow Software Ltd. — P2M

• Building scalable cross-platform mobile applications with Flutter
• Developing native Android applications with Kotlin and Jetpack Compose
• Integrating REST APIs, Firebase, authentication, and real-time services
• Creating responsive applications for phones and tablets
• Applying clean architecture and maintainable state management

ELITE Research Lab LLC — Remote

• Leading research in LLMs, NLP, Computer Vision, and Multimodal AI
• Designing model evaluation and benchmarking pipelines
• Conducting AI experiments and performance analysis
• Supporting research collaborators and publication workflows
• Contributing to technical reports and open-source AI systems

GitHub statistics are generated by third-party services. A card may temporarily become unavailable because of API rate limits.