Multi-source discussion tracker with in-memory TF-IDF clustering. Reddit via Arctic Shift (no API key), semantic grouping without embedding quotas, Groq for summaries.

The clustering threshold decision (0.25 cosine similarity) took longer than building the whole thing. Too high → everything is its own cluster. Too low → unrelated threads merge. Ship it and tune.

Describe an API in plain English, get working FastAPI code back. Runs against any OpenAI-compatible endpoint including local Ollama.

Useful for rapid prototyping where you know what you want but don't want to type boilerplate. Not production — treat it as a scaffold starting point.

Scrapes transcripts from an entire YouTube channel and packages them for NotebookLM import. No YouTube API key needed.

Built because manually adding 80 videos one by one is not something I was willing to do. Four hours to write, saves the problem forever.

AI incident analyst built on NestJS + Prisma + HuggingFace Inference Router. Reads logs, identifies anomaly patterns, drafts a root-cause summary. Gemma 4 challenge submission.

HuggingFace Inference Router was the interesting choice here — routes to the best available model without hard-coding a provider. Actual production would need tighter prompt guardrails.

Adds autonomous agent mode to an existing content SaaS — ~600 lines, no architecture rewrite. Hermes orchestration challenge submission.

The constraint of not touching the core SaaS codebase made this interesting. Grafted agent planning on top of existing endpoints rather than rebuilding from scratch.

Manual-trigger cross-poster from blog to Reddit. One command, one post. No automation, no scheduler — intentionally stays manual to avoid looking like a bot.

The right tool here is the dumbest one that works. Spent ten minutes writing it and it does exactly what it needs to do, nothing more.

SDK examples and integration recipes for the Idea2Post API — PHP, Python, Node. Topic → 10 content formats. Free tier: 3 requests/day.

Exists because "read the docs" only goes so far. Working code in your stack is always faster than translating curl examples.

Integration examples for the Idea2Viral video API. Covers job submission, status polling, download, and the optional auto-publish to YouTube flow.

The async job pattern is the part people always get wrong the first time — this shows the correct poll-then-download loop.

Working code examples for AIGate image and video generation API — covers Flux, SDXL, Kling, and WanVideo endpoints.

Image generation APIs all have slightly different async patterns. This normalises the differences so you can swap models without rewriting the polling logic.