Briefstamp: a receipt for the website handoff

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

A freelancer friend of mine once took a ₹8,000 website gig. After he delivered, the client made him add extra things that were never in the brief. He had nothing to point at to show what was actually agreed. That's the gap Briefstamp fills.

Briefstamp is the receipt for that handoff. Paste the client's brief, approve the checks it proposes, and a real browser tests the site. You get a report showing what matches the agreement, what fails, and what still needs a human call. An approved checklist doesn't guarantee payment, but it draws a clear line between what was agreed and what is extra.

The rule: the model only proposes checks. A human approves them, a real browser observes the site, and code decides pass or fail.

Who pays: freelancers and small studios who hand over websites. My pricing idea is ₹499 per handoff, a guess I still need to test.

Demo

Live app · Judge guide · MP4 backup

One real run, on October 16:

1. Brief: Six explicit homepage requirements: the exact heading "Send the site. Show the proof.", a delivery tagline, a "See the proof" link, image alternatives, mobile fit, and a subjective trust requirement.
2. Proposal: Gemma proposed six typed checks from those six requirements. The subjective requirement remained for human review. Nothing ran yet.
3. Approval: I approved the checks. Unapproved ones are rejected.
4. Audit: Chromium opened the real site: five passes, one unverified requirement.
5. Report: Audit results and screenshot references were saved to Tiger Postgres. The exported report included the actual screenshots and observations for the client.

To show the checks aren't rubber stamps, I changed the expected heading on purpose. The audit failed, as it should. Full receipt

Optional: public evidence agent

The main demo works without this. If you want to see the agent, open the evidence walkthrough, approve its six labelled template checks, run Chromium, then consent to the model/search step. Gemma picks bounded tools (saved failures from Tiger, a SerpApi search) and cites what they returned; it can't change browser results. No token needed.

Providers can fail: of four retained public attempts, one completed, two returned Google HTTP 500 and one hit a search timeout. The shared allowance is 100 runs total, six per hour and one per browser session, ending October 20, 2026 at 00:00 UTC, and failed reservations stay spent. Public receipts · Allowance readback

Code

Public MIT repository. The public judge source passed 963 app tests, 19 Python tests, the build and real browser audits. CI record

How I Built It

The model only proposes. It never writes a browser script and never decides pass or fail.

Gemma proposes typed checks through Mastra, I approve them, Chromium observes the site, and Tiger saves audit results and screenshot references; the downloadable report embeds the screenshots. A separate Temporal local/CI recovery demonstration resumed after a crash without repeating the browser run; it is not an always-on Render worker. A second native Gemma agent selects saved failures from Tiger and a SerpApi search, then cites actual returned facts. Backboard's protected proof recovered saved results. ElevenLabs generated a synthetic counts-only audio sample; it decoded successfully, but its audible wording has not been independently checked.

Why Does Open Innovation Matter?

Open weights let me run Gemma locally and fine-tune a small Qwen model for this one job, while keeping the browser runner independent of any model host. A freelancer's client data doesn't have to live with one AI vendor.

My First User

The friend with the ₹8,000 gig, Priyanshu, tried Briefstamp and told me he can pay for it. No price was named, and that's one person, not validation: I haven't tested a price with him or watched him use it on a real handoff. Notes

Prize Categories

Each entry below links its proof. Awards are the judges' call.

Best Use of Gemma

Gemma is the step that turns a client's plain-English brief into checks a browser can run. Hosted gemma-4-26b-a4b-it proposed six checks for the real Briefstamp homepage using 389 input and 296 output tokens; a real browser ran them only after approval. Proof

Best Use of Tinker

I LoRA fine-tuned Qwen3.5-4B to write these checks and compared it with a few-shot prompt on the same 24 requests. It used 69.55% fewer input tokens (16,737 down to 5,097). The tradeoff: it matched the reference 22/24 times versus 23/24, and it was slower. Runs and math

Best Use of Render

Render's free tier hosts the interface, the API and the isolated Mastra worker. A judge can approve an audit, see screenshots and download the report from the public URL. Deployment proof · Hosted desktop and mobile run

Best Use of Tiger Data

Tiger Data stores failure embeddings in a pgvector vector(384) column. One SQL statement combines keyword and cosine ranks using reciprocal rank fusion. "Sideways panning on smartphones" still ranks mobile overflow first: SQL cosine 0.378, matching an independent calculation, with an RRF score of 0.0164. pgvectorscale and a valid DiskANN index are installed; this bounded exact-ranking query does not prove index use or speed. Only three fixed query vectors are supported. Migration and scores

Best Use of SerpApi

An approved search_presence check matches the expected title against an exact authorized page URL. SerpApi returned three organic results with one matching owned-repository result at position 1; the demo search returned zero. The earlier profile mismatch and timeout remain retained. The positive target came from an earlier result, so this is a control, not an independent benchmark. Four dedicated reservations remain spent; no public judge slot was used. Empty results do not establish de-indexing or browser acceptance. Responses and receipts

Best Use of Mastra

A real Mastra workflow proposes, validates and suspends before browser execution. In CI, the same run resumed with six approved check IDs bound to its plan digest, then ran Chromium once: five passed and one remained unverified. It replayed an earlier actual Gemma proposal, with no new inference or judge spending. Approval came from an automated verifier, not a customer. This process-local proof does not establish durable storage, restart recovery or hosted public approval suspension. States, payload and screenshots

Best Use of Sentry Agent Tracing

Sentry showed where a freelancer's wait goes: a 23.3-second proposal, of which generation was only 7.6 seconds. Without the trace I'd have blamed the model for all of it. Trace

Best Use of GitHub Copilot (GitHub Actions)

My workflow audits a healthy site, a deliberately broken one, then the fixed one with the same approved plan: 5 passes → 4 failures → 5 passes, with screenshots and reports uploaded. Workflow · Run

Best Use of Temporal

Temporal wraps both Gemma turns, Tiger retrieval and SerpApi search as separate activities. Real CI recovered from an injected Gemma 500 and pre-dispatch search timeout, each with one retry. After SIGKILL following Tiger, three completed activities were not repeated. Permanent HTTP 400 failed after one attempt; persistent 500 failed after three. The retained-provider-fixture proof spends no judge allowance and does not claim an always-on hosted worker. Event histories and receipt

Earlier, I restarted both the Temporal server and worker mid-run; the workflow resumed and the browser ran exactly once. Recovery run

Best Use of Backboard

After an audit, Backboard stored the results as project memory, and a later search recovered all six matching facts. The idea: the next handoff for the same client starts from what was already checked. Receipt

Best Use of ElevenLabs

The idea: a client who won't open a report can hear the result. ElevenLabs returned a synthetic counts-only MP3 that independently decoded as 22.987755 seconds. Voice generated with ElevenLabs; synthetic demo only. Listening and audible wording remain unverified. Listen

Also used, not entered: none of Entire, .tech, Arduino, TabPFN, DigitalOcean or MongoDB Atlas has full evidence here, so I'm not claiming them.

What It Can't Do Yet

• The public demo audits only the Briefstamp homepage and a sample site, not any client URL. My approval isn't the client's acceptance; that's the next step.
• The video shows labelled templates. The real Gemma run is separate, dated evidence, and earlier attempts included an HTTP 500 and invalid JSON before the successful one.
• The Tinker model's training examples haven't been checked by a human, and the agent, search, speech and memory demos run on limited trial credits.

Full limitations and funding details

What's Next

Watch real freelancers use it on real handoffs, and find out whether ₹499 is a price they'd pay. If you freelance and want to try it on your next delivery, reach out.

AI helped with coding, testing and drafting. Recorded approvals include automated verifier/browser actions under my authorization; independent client acceptance is not claimed.

Related reading: Human-in-the-loop QA, metataro · Playwright post-mortem, tamizuddin