AI and Machine Learning for Quantitative Research

Building intelligent, reproducible systems for quantitative modeling, financial analysis, forecasting, and risk management.

I am pursuing an M.S. in Quantitative Methods with a concentration in Mathematical Finance, and my career goal is to become an AI Engineer in Quantitative Research. This portfolio presents reproducible projects combining machine learning, statistics, mathematical modeling, financial analysis, and production software engineering. Each project progresses from a clearly defined research question to tested code, documented results, containerized execution, and automated CI/CD — full source, tests, and READMEs are on GitHub.

Project roadmap

Project	Focus	Status
Numerical Methods for Quantitative Research	Lagrange and Neville interpolation, error, stability, Python and R — live demo in the next tab	Implemented foundation
Monte Carlo Option Pricing and Risk Analysis	GBM, Black–Scholes, confidence intervals, VaR — live demo in the next tab	Implemented
Financial Time-Series Forecasting	Baselines, walk-forward validation, leakage control	Planned
Machine Learning for Credit-Risk Prediction	Calibration, imbalance, fairness, explainability	Planned
Financial News and SEC Filing Analysis	Sentiment, topics, transformers	Planned

Important disclaimer

This portfolio is for education and research. It is not investment advice. Simulated or historical performance does not guarantee future results. A research model is never presented here as a profitable trading strategy without appropriate out-of-sample evidence, costs, and risk analysis.

Author: Lameck Nyakweba — M.S. Quantitative Methods, Mathematical Finance concentration; aspiring AI Engineer in Quantitative Research.

Appeal: Sample Workings to Augment an Earlier Submitted Assignment

On my last submission for this Neville's Method question, I lost marks because I only handed in the final answer with no working shown — there was nothing to check my process against. This page accompanies that submission with the full working, computed live rather than typed by hand, using the exact recurrence we covered in class:

Q[i, 0] = f(x_i)

Q[i, j] = ((x_target - x[i-j]) * Q[i, j-1] - (x_target - x[i]) * Q[i-1, j-1])
          / (x[i] - x[i-j])

Defaults below are the assignment's own data (x = 1.0 to 2.5, target x = 1.5, expected result f(1.5) = 0.5118276664). Every working can be checked two ways: directly on this page — the table and every substituted calculation below — or independently, using the CSV and Python downloads next to the result.

x-values (comma-separated)

f(x)-values (comma-separated)

Evaluation point (target x)

Download and check independently

CSV — not just a transcript: alongside the finished value, each row also has the recurrence's five numeric inputs in their own columns. Open it in Excel, build the formula yourself in a spare column from those columns, and confirm it matches — a ready-to-test solution, not only something to read.
Python source (neville.py) — the real code that produced every number above, not hand-typed workings. It needs only numpy and nothing else from this project, so it can be downloaded on its own and run/validated on any other machine: pip install numpy then python neville.py reproduces the table above from scratch.

Download the complete table as CSV

Python source (neville.py) — runs standalone anywhere

neville .py

11.2 KB ⇣

Complete recursive table

Monte Carlo Option Pricing and Risk Analysis — live demo

Every number below comes from the exact same quant_mc_options package used in the project's own test suite — this Space installs it directly from the GitHub repository rather than re-implementing the math. Inputs are simulated market parameters, not live or historical market data; see the project README for the full methodology, evidence classification, and limitations.

Large-N Monte Carlo on ZeroGPU

Same GBM pricing model as the CPU demo, reimplemented in torch to scale to path counts the CPU version isn't meant for (tens of millions), using a ZeroGPU allocation. This is a scaling/engineering demonstration, not a more accurate price -- the CPU Monte Carlo result already agrees with Black-Scholes to within its reported confidence interval at far smaller path counts.