sanity ♠️ retweeted
1 am.
Our first early game prototype.
Does it look like something you'd want to play?
#opengamejam
1 am.
- Following proper Game development life cycle for this one.
- Prototyped our antagonist, what vibe is the game giving?
Thanks to @MeshyAI for making things super easy
Face still needs work. He's not smiling for a reason. 🐐
#opengamejam
sanity ♠️ retweeted
Wanted to keep this to myself but nah let me post this here so y'all will congratulate me the same way I congratulate y'all 😂
Last week friday, I passed phase 1 of my $5k account @EquityEdgeUK 💚
From Draw down to making 10.4% a day 🔥
Indeed GOD CAN TEACH A MAN 🙏
sanity ♠️ retweeted
Day 2.
Still up generating models. img gen in 4k, model gen in 4k.
Idk if this game's gonne be heavy, but our aim is definitely premium quality.
Prototyping, not started.
#opengamejam
sanity ♠️ retweeted
United Senior & Ayangburen senior High School, lkorodu 10th Nov ✅ .
Agnus Memorial Memorial High school - 11th Nov ✅
Ojota senior secondary school - 17th Nov ✅
Ogudu Senior Grammar school - 17th Nov✅
Ipakodo senior grammar school - 18th Nov ✅
5/7 schools done.
sanity ♠️ retweeted
we will win.
not immediately, but definitely.
provided there’s consistency, growth is inevitable. 🖤
sanity ♠️ retweeted
i built lokvane for the @openservai hackathon #SERVHackathon
etsy sellers can ship products worldwide, but there's a problem with how their listings are translated.
etsy automatically translates listings into different languages, and sellers often have no idea what buyers are actually reading.
a simple translation mistake can change the entire meaning of a product.
imagine selling a 19-piece glassware set and having it translated as a set from the 19th century.
that's a real problem.
the solution: lokvane
i built lokvane, a localization agent that helps etsy sellers identify bad machine translations, correct them, and improve future listings using independent ai evaluations.
germany is our first supported market.
the idea is simple: every listing should benefit from what the agent learned while correcting the previous one.
how it works
a seller provides two things:
- the original english product listing.
- etsy's german auto-translation.
lokvane sends both to serv reasoning.
>serv reasoning compares them, identifies misleading phrases, explains why they're wrong, and generates a corrected german listing.
this includes a new title, description, and 13 local search tags.
everything is ready to copy back into etsy.
where reppo comes in
>generating a correction isn't enough. we need to know whether the correction is actually better.
this is where reppo comes in.
lokvane submits both versions for independent evaluation:
- etsy's original machine translation.
- lokvane's corrected version.
reppo evaluates both and returns scores, decisions, confidence levels, and written critiques.
the agent doesn't grade its own work.
after reppo evaluates both versions, lokvane takes the evaluators' critiques and feeds them back into serv reasoning.
serv reasoning extracts useful lessons and turns them into rules for the german market playbook.
every rule stores the reppo evaluation id that produced it.
the next listing uses those rules to generate a better correction.
fix → evaluate → learn → improve.
we tested the learning loop
i ran eight rounds against the live serv reasoning and reppo apis.
some results from repeating the same listings:
- vintage glassware: 80 to 100 with 11 learned rules.
- sterling silver ring: 76 to 94 with 12 learned rules.
- botanical wall art: 92 to 96 with 12 learned rules.
these are reppo evaluator scores, not sales figures. the evaluation panel can vary between runs, so the results aren't proof of actual sales improvement.
every evaluation id is stored and available for verification.
serv reasoning powers the entire generation process. it handles error detection, listing corrections, and the extraction of new market rules.
reppo provides independent evaluations and the feedback that powers the learning loop.
without serv reasoning, lokvane cannot generate its fixes or learn new rules.
without reppo, it has no independent evaluation or learning signal.
both are essential to how the product works.
what i built
the project includes:
-serv reasoning integration with strict json schema outputs.
-reppo evaluation api integration.
-automated evaluation and verdict polling.
-a market-specific learning engine.
-a persistent playbook and round history.
-a complete web application.
-api request guards, rate limits, and daily usage budgets.
-unit tests, type checking, and ci.
-production deployment with docker and dokploy.
you can explore the product, run your own listing through the tool, inspect the learning history, and trace individual rules back to their reppo evaluations.
with lokvane, i wanted to take that idea into a practical problem that affects people selling products online.
every correction produces feedback. every useful piece of feedback can become a rule. and every new listing gets to benefit from what came before it.
product: lokvane.midelabs.xyz
try the tool: lokvane.midelabs.xyz/check
source code: github.com/mystiquemide/lokv…
here is a 1 minute demo video walkthrough of how lokvane works






























