Independent software developer focused on local AI architectures, media automation, data visualization, and game tooling. I build self-contained, high-performance applications designed to run locally on consumer hardware without subscription tiers, telemetry, or external API lock-in.
ko_fi: bloopieblair https://ko-fi.com/bloopieblair
Local-first autonomous deep research and intelligence studio
LocAi Search is a desktop research workstation that runs small, open-weight language models locally on consumer hardware while performing live web research and source extraction.
- Local Inference: Directly interfaces with llama.cpp and Ollama to run GGUF quantized models (such as Qwen2.5-7B-Instruct) with CUDA acceleration and zero external AI telemetry.
- Neural Semantic Re-Ranking: Embedded FlashRank cross-encoder (
ms-marco-MiniLM-L-12-v2) scoring and ranking extracted web documents against queries using semantic relevance and domain authority. - Epistemic Anti-Hallucination Guardrails: Deterministic code-level gates enforce factual grounding, including a claim-to-source digit token linter, symmetric entity coverage requirements for comparison queries, and extraction validity filters.
- Desktop Architecture: Native desktop GUI built on PyWebView, FastAPI, and WebSockets for real-time token streaming and tool execution tracking.
Stack: Python, llama.cpp, FlashRank, Playwright, FastAPI, PyWebView, Hugging Face Hub
Automated YouTube Shorts generator and video production pipeline
An end-to-end media automation pipeline and web studio that extracts top Reddit discussions and renders fully composited 9:16 vertical video Shorts for YouTube.
- Content & Card Extraction: Dual-engine Reddit scraper that renders pixel-perfect Reddit card UI components into transparent PNGs using headless Chromium (Playwright) or fallback Pillow graphics.
- Speech Synthesis: Converts story text to voiceover using Microsoft Edge neural TTS, with pre-flight sanitization for spoiler tags, links, and pronunciation anomalies.
- Hardware-Accelerated Compositing: Three-tier rendering engine with native FFmpeg hardware encoding (NVIDIA NVENC, AMD AMF, Intel QSV) and automated fallback to MoviePy and OpenCV.
- Dynamic Captions & Publishing: Synchronizes word-level animated subtitles with voiceover timestamps, crops background gameplay to 9:16 vertical orientation, mixes procedural lofi audio, and automates uploads via YouTube Data API v3 OAuth.
Stack: Python, FastAPI, Playwright, Edge-TTS, FFmpeg / NVENC, MoviePy, OpenCV, YouTube Data API v3
Interactive synergy graph and deck analysis suite for Magic: The Gathering
A browser-based analytical toolkit and visualization engine for Commander / EDH deckbuilders to inspect card mechanics and simulate early-game consistency.
- Synergy Network Graph: Renders decklists as interactive D3.js force-directed graphs, mapping mechanical relationships (ramp, card draw, removal, tutors) with colored multi-link paths and isolated node detection.
- BFS Goldfish Simulator: Simulates opening turns using a breadth-first search pathfinder across branching game states to evaluate play lines and commander cast timing.
- Statistical Dashboard: Interactive converted mana cost distributions, land production vs. pip requirement ratios, and hypergeometric probability calculators for opening hands.
- Scryfall & Arena Integration: Live Scryfall card data resolution, high-resolution card artwork, and MTG Arena collection log synchronization.
Stack: TypeScript, React 18, Vite, D3.js, Scryfall API, TailwindCSS
- Languages: Python, TypeScript, JavaScript, SQL, HTML/CSS, Bash, PowerShell
- Backend & Systems: FastAPI, Uvicorn, WebSockets, REST APIs, Pytest, Pydantic
- AI & Local Inference: llama.cpp, Ollama, GGUF quantization, FlashRank, Hugging Face
- Frontend & Data Visualization: React, Vite, D3.js, TailwindCSS, PyWebView
- Media & Browser Automation: FFmpeg (NVENC/AMF/QSV), MoviePy, OpenCV, Playwright, Pillow, Edge-TTS
- Platforms & Tooling: Git, Linux, Windows, Docker
- GitHub: @BloopieBlair
- Repositories: github.com/BloopieBlair?tab=repositories