AI Agents

LLM-driven agents I'm building

Early-stage concepts — where an LLM has to do more than answer a question, it has to decide and act.

PCPilot

Concept

A natural-language assistant that runs entirely on your machine via a locally-hosted Ollama model. Ask it to find files, clean up disk space, check what's eating CPU or memory, or explain what's slowing your laptop down — no cloud calls, nothing leaves the device.

The idea is to point a local LLM at a small set of well-scoped system tools — file search, process inspection, disk usage — and let it turn something like 'why is my laptop slow' into the right sequence of read-only checks before it's ever allowed to touch anything destructive.

Ollama
·Local LLM
·Tool Calling
·System Automation
·Python

PR Sentinel

Concept

An agent that watches open pull requests, reasons about a diff in the context of the surrounding code, and leaves inline review comments — flagging correctness issues and simplification opportunities before a human reviewer looks at it.

Less interested in another linter and more in something that reads a change the way a careful reviewer would: what does this break, what's now redundant, what's missing test coverage — then says so, inline, instead of a wall of text.

LLM
·GitHub API
·Static Analysis
·Agentic Workflows

Draftsmith

Concept

Given a topic, pulls and cross-checks sources, then produces a structured first draft with citations — meant to compress the research phase of writing, not replace the editing and judgment that comes after.

Most of the time I spend writing isn't the writing, it's gathering and reconciling sources. The goal here is an agent that does that legwork and hands back a draft I still have to rewrite in my own voice, not one I publish as-is.

LLM
·RAG
·Web Search
·Content Generation

SignalDesk

Concept

An agentic layer on top of the existing SwingTrading Automation pipeline — instead of just outputting a buy/sell signal, it reasons in natural language about position sizing and risk, and can explain why it flagged or skipped a trade.

The screening and signal generation already exist; what's missing is the reasoning layer that can justify a call the way a human trader would, so the output is an explanation, not just a number.

LangChain
·Alpha Vantage API
·LLM Reasoning
·Python