I Built an App That Hunts Gigs While I Sleep

Odemakinde Ezekiel Temitayo — Sun, 25 Oct 2026 21:02:24 +0000

My DEV Hacktoberfest Week 3 entry: LazyGig — an overnight gig scout powered by open-weight AI.

Every freelancer knows the dread. You wake up, and before coffee, before anything, there's the ritual: open the gig boards. Refresh. Scroll. Twenty listings, nineteen of them $0.05 microtasks or "translate this into Portuguese" or referral spam dressed as opportunity. By the time you find the one real gig, someone else already applied.

I hated that ritual. So for Week 3's "Slow and Steady" theme — build something that rewards laziness — I automated the whole miserable morning.

Meet LazyGig 🌙

LazyGig (repo) watches freelance gig boards overnight and hands me a morning brief: gigs scanned, genuine matches, and a drafted application for each one. I tap Approve on the drafts I like. That's the entire job now.

Here's what happens under the hood when I hit "Run overnight scan":

Ingest. It pulls live listings from two real boards — Superteam Earn's agent endpoint (I'm a registered agent there) and pact0's public job board. No mocks; if the boards are empty, it says so.
Rank. One batched call to Llama 3.1 8B — open weights, running on Cloudflare's Workers AI free tier — scores every gig 0–100 for genuine skill match and rate fairness, with a one-line "why it fits."
Draft. A second batched call writes a short, concrete, flattery-free application for each top match.
Decide. The dashboard shows ranked cards with score bars and the drafts. Approve or Skip. Approving just marks it — nothing is ever sent automatically. Demo-safe by design.

The part I'm proudest of is the ranking prompt's honesty. I told the model to be brutal: $0.05 microtasks and off-profile work score low, even if they're vaguely doable. On its first real run it scanned 13 gigs and surfaced exactly one match — an 80/100 — with a draft ready. Twelve rejections, zero wasted mornings. LazyGig would rather show me nothing than waste my time, and that sentence is basically the product spec.

Why open weights made this possible

Here's the thing nobody says out loud: I couldn't have built this two years ago. Not for zero dollars.

LazyGig runs its entire brain on an open-weight 8-billion-parameter model served from Cloudflare's free tier. Total AI spend: $0. Total infrastructure spend: $0. I'm a builder in Nigeria; I don't have a corporate AI budget or a GPU cluster. Open weights mean the frontier isn't a place I read about — it's an API I can actually call.

That's what open innovation means to me, concretely: the gap between "I have an idea at 11pm" and "it's live and working" is now one evening, not one funding round. Llama's open weights, Cloudflare's free Workers AI, public job-board APIs, a MIT-licensed repo anyone can fork — every layer of LazyGig is something I didn't have to ask permission to use. When the tools are open, the only moat left is who ships. And shipping is the part I can control.

So here's to slow and steady. While I sleep, the scout works. While the scout works, the weights are open. Everybody eats. 😴

Built with: Cloudflare Workers, Workers AI (@cf/meta/llama-3.1-8b-instruct-fp8), Cloudflare KV, vanilla JS. Live at lazygig.odemakindeezekiel5.workers.dev · Code at github.com/nostoryworld/lazygig.

devchallenge #hf26challenge

DEV Community: Odemakinde Ezekiel Temitayo

I Built an App That Hunts Gigs While I Sleep

Meet LazyGig 🌙

Why open weights made this possible

devchallenge #hf26challenge