kemen

AI-Engineer | Python Developer | Researcher

[kemen @kemencode]

@igor_mikerin @Polymarket @PolymarketTrade It's not a bad strategy, I haven't tried it yet. How you researched it?

[kemen @kemencode]

The best approach to choosing vibe-coding tools

I see so many people obsessing over which AI coding tool is "the one". They're looking for a magic button that solves everything. Spoiler: it doesn't exist. LLMs aren't there yet. But your approach can make up for it.

—

Before I share my stack — here's what actually matters.

My dev criteria: — Speed — Interface UX — Code quality — Token cost
Without understanding the process, you'll always be trading one for another.
—
My current stack: Claude, Claude Code, VS Code, Replit, Kiro
Phase 1 — Research + Design: Claude chat (Opus 4.6) — finding the optimal technical solution and generating docs for the coding agent. This is the brain. Before writing a single line of code, I squeeze maximum context and structure out of Opus.

—

Phase 2 — Execution:

VS Code + Claude Code — heavy lifting. Bug fixes, financial modules, complex logic. Surgical, powerful moves.

Replit — great for MVPs. Builds solid UI fast. You can demo to a client in a day.
Kiro — scaffolding apps from a tech spec. Feed it the spec — get a skeleton to build on.

—

Ratings (out of 10):

Claude chat — Speed: 9 | UX: 8 | Research quality: 8 | Cost: $20/mo sub, more than enough for research

VS Code + Claude Code — Speed: 7 | UX: 8 | Code quality: 8 | Cost: on heavy tasks the $20 limit hits fast. Use it surgically.

Replit — Speed: 6 | UX: 8 | Code quality: 7 | Cost: absurd. No comment.

Kiro — Speed: 7 | UX: 8 | Code quality: 7 | Cost: free. Yes, free.

—

Key takeaway:
There's no perfect tool. There's a perfect approach.
You can grab the most hyped, most expensive tool on the market — but without knowing when and why to use it, you're just burning tokens. Meanwhile someone with a free stack and a clear process is shipping faster than you.
A tool without an approach is just a fancy way to burn tokens.

—
If after reading this your question is still "which tool is the best" — read it again.

[kemen @kemencode]

@cryppimagic product looks good, keep going brother!

[kemen @kemencode]

@aashatwt Link please!

[kemen @kemencode]

@xmayeth genuinely curious what execution speed he's getting. py-clob-client adds like 200-300ms latency and on 5min markets that seems brutal

[kemen @kemencode]

@phosphenq @Polymarket @PolymarketTrade $1.5-5k per trade on a $124k bankroll is aggressive. one choppy BTC week and that curve gets ugly fast

[kemen @kemencode]

@ongodmarkets1 sounds interesting. I know thats you can suggest new market in Polymarket's discord.

[kemen @kemencode]

11,603 trades. $118M volume. $645,489 profit.

This Polymarket trader made more in 2 months than most devs make in 5 years.

I almost scrolled past this account. The username looked like a throwaway. But the numbers made me stop.

Account88888 joined Polymarket on December 4th. Two months ago.

Since then:
→ $118,705,215 in total volume
→ 11,603 trades executed
→ $645K in realized profit
→ Largest single win: $62,860

That's roughly $9,200 per day. Every day. For two months straight.

This isn't gambling. This is infrastructure.

The math tells the story:

Average profit per trade: $55.6. Not life-changing. But multiply that by 166 trades per day and you get a salary most people dream about.

$118M in volume from one account means this trader is moving serious size. The volume-to-profit ratio (0.54%) suggests market making or liquidity provision — not directional bets.

They're not predicting outcomes. They're capturing spread on both sides and letting volume do the work.

Strategy breakdown:
→ High-frequency execution (166+ trades/day)
→ Large position sizes (~$10K per trade average)
→ Consistent small edges stacked over thousands of trades
→ Likely automated — no human clicks 166 times daily for 2 months

The boring truth about Polymarket whales: they don't pick winners. They build systems that print whether markets go up or down.

Receipts:
> https://t.co/Ne2J50NXfB

$645K in 60 days. From what looks like a burner account.

What do you think — is high-frequency market making the real edge on Polymarket?

[kemen @kemencode]

@xmayeth bro you’re research machine!

[kemen @kemencode]

@Valera51063845 yeah, we're in game!
0.1$ x 100 = 10$

[kemen @kemencode]

605 trades. $337K volume. $28,498 profit. Not a single win over $1,554.

This Polymarket trader makes money the most boring way possible.

ClawdBot flagged this account last night and I almost skipped it because nothing looked impressive...

No massive bets. No 10x screenshots. No viral moments.

Then I looked at the numbers.

According to his profile:
> https://t.co/Ot7txlPado

$2,400 per month. Every month. Like clockwork.

His average profit per trade: $47. Forty seven dollars. Six hundred and five times.

The strategy:
→ Never bets big on a single outcome
→ Spreads across markets where he has an edge
→ Lets probability do the work
→ No emotions. No FOMO. No revenge trades

While everyone on crypto Twitter chases the next 100x, this guy makes $47, closes the tab, and does it again tomorrow.

$14,606 currently sitting in active bets. He is not done.

His largest win was $1,554. Most people would not even screenshot that. But stack 605 of those small wins and you get almost $30K.

This is the trader version of the guy who puts $500 into an index fund every month and retires early while his friends blow up chasing Tesla options.

Most traders lose money trying to be right once in a big way. This guy makes money being slightly right hundreds of times.

The boring traders are the ones actually making money.

Would you rather have one 100x bet or 605 wins at $47 each?

[kemen @kemencode]

@igor_mikerin @Polymarket @PolymarketTrade cool research bro! keep growing!

[kemen @kemencode]

@brelgino @MetaMask thank you for useful advices!!

[kemen @kemencode]

@Valera51063845 nice work man!

[kemen @kemencode]

@brelgino good article, man! thanks for sharing data and insights of your content process.

[kemen @kemencode]

@0xFiseo @claudeai facts!

[kemen @kemencode]

30k+ lines of code in a week, solo?

With vibe-coding — absolutely.

// context

I’ve wanted to write about this for a while. Trying to fit my full take on this phenomenon (which I genuinely believe will reshape the IT industry) into one post makes no sense — so here’s my personal experience.

I’ve been using LLMs for codegen for over a year now. I honestly can’t imagine going back to writing everything by hand. Not because I forgot how to code — I just don’t see a reason to do it manually anymore.

// pipeline

> Model selection — my current setup:
- Claude Opus 4.5 — for specs
- Sonnet 4.5 — for coding
> Generate multiple specs + manual edits:
- requirements = what to build?
- tech implementation = how to build it?
> Code review — my own eyes + fixes through chat
> Testing (after checking tests) → refactor/debug with chat
> Iterate until shipped

// results

> Product development speed: easily ~2x
> Less manual grind
> Downsides: slight quality trade-off (worth it imo)

If you’re curious, I can break down my entire workflow + tools I use daily for different tasks.

Interested? Drop a like and comment.

[kemen @kemencode]

@stanisloou @glider__ @BrianInCrypto @_biggids @Dev33187 It's cool new! congrats man!

[kemen @kemencode]

@maxyatsuk @preddytrade @PolymarketBuild congrats!

[kemen @kemencode]

@igor_mikerin @Polymarket @PolymarketTrade it's insane results for such small time! good job bro