Christopher
Beaulieu.

An open-source web app for managing Raid: Shadow Legends clan siege assignments — self-hostable anywhere Docker runs.

Drag-and-drop assignment board with an auto-fill algorithm, a real-time validation engine, PNG battle card generation via Playwright, and Discord DM notifications. Reference deployment runs on Azure Container Apps behind Discord auth.

An LLM-powered job search pipeline that fetches listings from 9 sources, scores each description against your skills profile, and surfaces ranked results in a local web UI.

Multi-provider scoring (Claude, GPT-4, Gemini) with cost-per-call awareness and provider failover. Data stays local. One external contribution merged — a DB query optimization from a community fork.

A terminal-based AI agent that triages support tickets across HackerRank, Claude, and Visa product corpora. Built solo for HackerRank Orchestrate (May 2026, 24 hrs); judged 65th of 12,885 participants across 1,349 submissions.

Nine-stage RAG pipeline — preprocess → safety → router → query expansion → classifier → retrieval → product-area resolution → abstain → generate — with per-stage unit tests, structured run tracing, and explicit abstention on out-of-corpus or high-risk tickets. Multi-provider LLM client with deterministic outputs.

A 4-iteration empirical study quantizing Qwen3.5-0.8B for an agent on a CPU-only consumer-class envelope (8 threads, 16 GB).

Falsified the bandwidth-bound prediction at Iter-1, validated a refined dequant-overhead mechanism at Iter-2, and confirmed a novel KV-noise-as-attention-regularization hypothesis via a directional series. Recommended config: +18–27% decode, −16–31% memory, no quality regression.