Waqas Bin Hamed

Currently building data pipelines and ML systems at The Home Depot. Computational math background from Waterloo — I care about how algorithms work, not just that they do.

Personal & Side Projects

Spotify Playlist Organizer  [Live App]

Stack: Gemini API, Spotify API, LangChain, Python, Google Cloud Run     2025 – Present

An agentic LLM classification pipeline that automatically organizes Spotify libraries by mood and genre. Built with Gemini and LangChain using multi-step reasoning with structured output (JSON schema enforcement) to ensure reliable, consistent categorization. Deployed on Google Cloud Run with live user testing.

Access is gated via Spotify’s developer authorization — reach out to be added as an authorized user.

Key techniques: agentic tool use, structured output, prompt engineering, REST API integration, cloud deployment

Sum-of-norms regularized NMF [GitHub] [Paper]

Stack: Python, NumPy, SciPy     Published in Neural Computation (2026)

An optimization algorithm that automatically discovers the correct number of latent components in high-dimensional data — solving the rank estimation problem that limits standard NMF. Learns sparse, interpretable structure from data with applications in signal decomposition, recommendation systems, and scientific computing.

Key techniques: convex optimization, regularization, ablation studies, numerical benchmarking

Nonnegative Unimodal Matrix Factorization [GitHub]

Stack: Python, NumPy

Python implementation of NuMF, enforcing single-peak (unimodal) constraints on factorized components — useful for spectral analysis, bioinformatics, and signal processing where underlying patterns have natural peak structure.

Key techniques: constrained optimization, scientific computing, numerical methods

Professional Projects

SYMX.AI X.Parts

Organization: SYMX.AI (formerly Symboticware)

An NLP system that transforms unstructured maintenance logs into structured, queryable data — enabling fleet-wide fault analysis to reduce equipment downtime and fuel costs.

My contributions: Fine-tuned a DistilBERT classifier on domain-specific mining logs (~500 → ~5,000 training pairs via NLP augmentation), improving fault classification from ~50% to ~75% accuracy. Took the system from internal PoC to a shipped commercial product.

Intellistocks.ai

Organization: Big Byte Insights

A financial analytics platform extracting investment signals from SEC filings, earnings calls, and social media using NLP and predictive modeling.

My contributions: Built NLP ingestion pipelines processing 10+ GBs/day via REST API integrations; contributed to fine-tuning a financial sentiment classifier using HuggingFace Transformers.