Cybiqon Lab

Notes from the workshop

What we are building, what broke, and what the numbers said. Written by hand, published when there is something worth saying — not on a schedule.

entries 13since 2026-07-28last 17 Oct 05:30 IST

date
words
7 394
read
37 min
sources
13 sources
videos in 10 days
40 videos in 10 days
voice cost
$0 voice cost
quality gates
40 quality gates
voice languages
2 voice languages
Claude Code per run
$11–26 Claude Code per run
channels posted by cron
2 channels posted by cron
views
16 views

We deleted the voice bill. Now the studio posts its own videos.

Our open-source Claude Code video studio now clones my voice offline in English and Hindi, renders on a CPU and posts to YouTube and X by itself: 40 videos in ten days, $0 for voice. Instagram is still by hand, and the post says why.

AI Video · Claude Code · HyperFrames

date
words
6 889
read
35 min
sources
10 sources
finished videos
12 finished videos
HeyGen spend
$8.20 HeyGen spend
per voice minute
~$0.14 per voice minute
quality gates
37 quality gates
longest render
19.5 min
views
8 views

Twelve videos in four days. Nobody picked up a camera.

An agent now researches, scripts, voices and renders our videos with HyperFrames and a HeyGen clone of my voice: 12 finished in four days for $8.20 of voice. The first was 'too fast', the next 'too expert', then 'looked alike'. Every complaint became one of 37 pass/fail gates.

AI Video · HyperFrames · HeyGen

date
words
6 408
read
33 min
sources
11 sources
windows
25 → 169
builds
15 → 37
artisan scenes
72 artisan scenes
play review
15 days
missed by
3 days
views
5 views

We built the game in six days. The store review took fifteen.

We built ChitraYatra in six days — 25 windows to 169, five ways to play, 72 AI-generated scenes and four generated scores — and then Google Play's review took 15 days and we missed RevenueCat Shipaton by three. This is how the whole game was made, including the rule we broke on day two.

Games · Shipaton · Flutter

date
words
6 151
read
31 min
sources
9 sources
tests
0 → 171
review
1 round
idle
4 months
package
1.2 MB
sourcemaps
−2.1 MB
guard misses
1/30 guard misses
views
24 views

Our Chrome extension sat untouched for four months. Review took one day.

MapWit is a free Chrome extension for Google Maps lead generation: it scores the local businesses in a search and exports them to Excel, with no server and no account. It sat untouched for four months, went from zero tests to 171 in one day, and passed Chrome Web Store review on the first round.

Chrome Extensions · Engineering · Security

date
words
7 670
read
39 min
sources
36 sources
windows
25 → 169
generated
17 generated
collections
5 → 18
gcd grids
81/81 gcd grids
shipped
3 Oct
views
7 views

Ten Indian art traditions made the mood board. Two made the game.

Our Shipaton 2026 puzzle game uses kolam, rangoli and mandala, all generated by our own code. A kolam is a mirror curve: an m×n dot grid draws gcd(m, n) lines. Gond, Madhubani and Warli wait for commissioned artists, and deities and yantras stay out.

Games · Shipaton · Indian Art

date
words
4 694
read
24 min
sources
5 sources
running cost
₹0/mo running cost
homepage weight
3.7 MB → 233 KB
tests
313 tests
transforms used
60 / 5,000
worker startup
7 ms
D1 tables
17 D1 tables
views
67 views

Their shop costs ₹0 a month. A finished feature stays off to keep it that way.

We built a client's dry-fruits store on Cloudflare's free tier — storefront, admin panel, real payments. It costs them ₹0 a month to run. The order-alert emails are finished, tested and deliberately switched off, because Cloudflare's email sending needs the paid plan.

Cloudflare · Ecommerce · Workers

date
words
7 601
read
39 min
sources
10 sources
levels
500 levels
in band
419/500 in band
bot vs human
r +0.14
verbs
8 verbs
tests
894 tests
prototypes
10 prototypes
views
14 views

Eight bots played all 500 levels. They were measuring the wrong game.

We built eight bots to play a 500-level arcade game and measure how hard each level is. They played it 720 pixels wide. Phones are 412. Levels do not scale, so the shipped game was roughly twice as hard as the game being measured: 172 of 500 levels in band, not the 421 we were reporting.

Games · Procedural Generation · Testing

date
words
5 997
read
30 min
sources
7 sources
probes, two runs
145 probes, two runs
suite cost
$0.105 suite cost
identical replies
99/114 identical replies
distinct replies
10 distinct replies
indirect leaks
0/28 indirect leaks
cost doc drift
33x cost doc drift
views
11 views

Our agent passed every red team probe. That was the problem.

We pointed a generated red team at our agent and it passed everything. Then we counted the replies: 99 of 114 were byte-identical. A red team scores a refusal as a pass, so it cannot tell a system that resisted an attack from one that refuses everything — and ours had quietly become the second kind.

AI Agents · Evaluation · Security

date
words
4 749
read
24 min
sources
7 sources
scenarios
41 scenarios
runs
123 runs
suite cost
$3.06 suite cost
undercount
4.5x undercount
untested tools
6/17 untested tools
views
19 views

Six of our agent's seventeen tools had never run.

Six of seventeen agent tools had never once run in production, including both of the ones that unlock a contact and charge for it. This is the harness that finally tested them — a real model in a completely faked world, 41 scenarios, 123 runs, $3.06 — and the cost blind spot it uncovered on the way.

AI Agents · Evaluation · LLM