From elliptic-curve cryptography to AI systems that solve olympiad mathematics — the throughline is verification-driven problem-solving.
Competed in Kaggle's AIMO3 as Team Proof Engine, alongside engineering lead Samuel Koh, building a Parallel Self-Consistency Pipeline for Olympiad-Level Mathematical Reasoning. The system ran GPT-OSS-120B across 8 parallel reasoning attempts per problem, paired with Python-based tool-integrated reasoning, and combined into a final answer using an entropy-weighted confidence vote — under a strict 9-hour GPU budget with binary, no-partial-credit scoring.
My role centred on the mathematics: diagnosing early reasoning bugs, designing verification-driven few-shot prompts, writing dual-verification guardrails, and manually solving reference problems to validate the pipeline's output.
Watch the full pipeline animation →Solved all 10 puzzles, including the Metapuzzle, in Harvard's annual international logic and combinatorial reasoning competition — open to participants from 166+ countries. The competition tests formal reasoning and systematic logical deduction under timed conditions.
Extends an efficient ECC-based signcryption scheme into a generalized construction that operates in signcryption, signature-only, or encryption-only mode as required — with full analysis against chosen-plaintext, chosen-ciphertext, forgery, and man-in-the-middle attacks.
Derives a series solution for steady, isothermal channel flow of a non-Newtonian Eyring-Powell fluid, with velocity profile, flow rate, and stream function obtained via the Adomian Decomposition Method.
Multi-agent conversational platform (MCP Agents, Groq, LLaMA-3.3-70B, FAISS semantic search) built among 6,247 global participants.
GitHub · Certificate →XGBoost behavioral classifier with real-time empathetic interventions, built in a 48-hour global competition — FastAPI + Node.js multi-agent pipeline.
GitHub · Certificate →End-to-end regression pipeline benchmarking five models; 88%+ R² with 5-fold cross-validation, deployed live on Hugging Face Spaces.
GitHub · Live app →Empirical regression quantifying income, structural, and demographic influences on property prices (R² = 0.92).
GitHub →Analyzed a 1,470-employee dataset (16.2% attrition); interactive dashboards identifying highest-risk cohorts.
GitHub →11 structured SQL queries extracting revenue patterns, genre preferences, and regional customer segmentation.
GitHub →For teaching, research collaboration, or PhD-related conversations — reach out directly.
sohaibhassan199@gmail.com