Personal projects · Data & Backend

Data tools and AI-agent tooling I build in Python.

A personal collection of projects: data pipelines, backend APIs, and MCP servers that give AI agents reliable access to real data. Built to keep learning, from a molecular-diagnostics lab background.

See my work

Focus

What I work on

Turning messy, real-world data into something software (and AI agents) can actually use, reliably and reproducibly.

Data pipelines

Pulling data from different sources and shaping it into clean, reliable datasets.

Backend & APIs

Python services and APIs for ingesting, storing, and serving data.

AI-agent tooling

MCP servers where the model decides and tested code does the reading, so values are never guessed.

Work

MCP tools for AI agents

A set of MCP servers that give AI agents reliable access to real, messy data. All published on PyPI and listed in the official MCP registry. The idea throughout: the model decides what it wants, tested code does the actual reading, so values are never guessed.

PyPI · MCP registry

excel-agent-mcp

Lets an agent read real messy Excel files: multiple sheets, auto-detects the header row under title rows, forward-fills merged cells. pip install excel-agent-mcp

code pypi

PyPI · MCP registry

pdf-agent-mcp

Pulls text and tables out of PDFs (invoices, reports) as clean rows instead of a flattened blob. pip install pdf-agent-mcp

code pypi

agentic-csv-cleaner

Cleans messy CSVs where the LLM only picks which cleaning steps to run and tested code does the transforms. pip install agentic-csv-cleaner

code pypi

price-data-mcp

Gives an agent live market prices and candles, so it can answer "what's the price right now?" instead of guessing from stale data. pip install hyperliquid-price-mcp

code pypi

Other projects

Live demo

Real-Time Market-Data Platform

Real-time ingestion → storage → OHLCV aggregation → live dashboard, with analytics (volatility, anomaly detection). Idempotent, tested, Dockerized.

▶ live demo code

Product-Catalog Scraper

Dependency-light scraper: pagination, retry/backoff, polite rate-limiting, clean CSV output. Adaptable to any site.

code

ML Validation Pipeline

Honest out-of-sample validation (NumPy): walk-forward, permutation nulls, to check whether a result is real or just noise.

code

About

Lab background, now in code

Before software, I spent six years in molecular diagnostics labs: sequencing, assay validation, and work under GMP. That's where I got careful with data: check it, reproduce it, don't trust a number you can't back up. I work in data and backend engineering now, and the lab background means I understand the domain as well as the code.

Contact

Say hi

Always happy to chat about data, backend, or AI-agent tooling. You can find me on GitHub or reach me at wesseltl@gmail.com.