Ayan Aggarwal

Designing AI‑native consumer products · 0→1, design systems, agents.

Prev. @ZeptoNow @OlaElectric @getsimpl @TheJupiterApp

[Ayan Aggarwal @bot_weiser]

@sdhilip Any experience using FluidVoice with Hindi / Hinglish or other Indian vernacular accents, especially code-mixed sentences?

[Raghav @lifeofraghav19]

🚨Launching Depth (building in public)

I've been tracking my body obsessively for 18 months.

Wearables. CGM. Blood-work. Lost 25kgs. 3x Hyrox. Multiple marathons.

I still couldn't answer one basic question: what's actually working?

So I built something. 🧵

[Om Patel @om_patel5]

SOMEONE BUILT A SINGLE CLAUDE.MD FILE THAT FIXES EVERY BAD HABIT CLAUDE CODE HAS AND IT HIT 78.5K STARS

it's based on andrej karpathy's public observations about how LLMs write code

the problem he pointed out is that claude makes silent assumptions, overcomplicates everything, writes 1000 lines when 100 would do, and sometimes deletes code it doesn't fully understand as a side effect

so forrestchang turned karpathy's critique into 4 behavioral principles and dropped them in one claude.md file:

1\ surface your assumptions

don't pick an interpretation silently. if there are multiple ways to read the task, say so. if uncertain, ask. push back when something doesn't make sense instead of just running with a bad plan

2\ minimum viable code

no speculative features, no abstractions for single use code, no "flexibility" you weren't asked for.

if you wrote 200 lines and 50 would work, REWRITE IT. ask yourself if a senior engineer would call this overcomplicated

3\ surgical changes only

don't touch code you don't fully understand, don't refactor unrelated stuff as a side effect, don't delete comments because they look unnecessary. only change what the task actually requires

4\ goal driven execution

give claude success criteria instead of step by step instructions.

karpathy's exact quote: "LLMs are exceptionally good at looping until they meet specific goals. don't tell it what to do, give it success criteria and watch it go"

one file has 78.5k STARS AND 7.4k FORKS on a single github repo.

install is one curl command that drops it straight into your ~/.claude folder

[Ayan Aggarwal @bot_weiser]

Test: sent it a viral AI video claiming MH370 disappeared through a wormhole.

TruthLayer replied in seconds:

❌ False. An Einstein-Rosen bridge is a purely theoretical concept in physics. No scientific evidence they exist. Creating one remains science fiction.

Grounded. No hallucinations. Telegram live. WhatsApp next.

[Ayan Aggarwal @bot_weiser]

Four weeks ago I started building a fact-checker for where misinformation actually spreads in India: WhatsApp forwards and Instagram Reels.

Not another text wrapper. A full multimodal pipeline.

Today, Phase 1 is live on Telegram.

[Ayan Aggarwal @bot_weiser]

TruthLayer handles 9 content types:

Text forwards
Images with captions
Images with text overlays
Screenshots of fake articles
Voice notes
Video with speech
Video with music + text overlays
Links

One verdict. Usually under 8 seconds.

[Ayan Aggarwal @bot_weiser]

Nobody questions a voice note from family.

That's why they spread faster than screenshots.

Sent a 9-second clip to TruthLayer. Someone claiming steam inhalation prevents coronavirus. Sounds like a doctor.

❌ False
WHO & CDC don't recommend it for prevention. It's just a home remedy for symptom relief, nothing more.

Text. Images. Voice notes.
TruthLayer now handles every format your family groups throw at it.

What's the most believable false forward you've received? Drop it, I'll run it live.

[Andreas Storm @avstorm]

Apple Korea has uploaded a new iPhone 17 Pro ad

[Ayan Aggarwal @bot_weiser]

Maybe the same. Ever since I started building, I reach for code before Figma.

But I don’t think the designer side dies. The taste, the pattern recognition, and the “this just feels wrong” sense are still running in the background. That’s what separates pure logic from actual design thinking.

Maybe that’s the part that makes us a bit special.

[Ayan Aggarwal @bot_weiser]

Every Indian family WhatsApp group has this right now.

Mom forwards a Reel: "Wear a copper ring, arthritis gone + stomach clean." 15,900 likes. Bold Hindi text. Looks legit.

Sent the screenshot straight to TruthLayer.

No typing. No translation prompt.

❌ False

Copper rings = placebo for joint pain. (The real Ayurvedic tip is drinking water from copper vessels, not wearing a ring.)

Text was easy last week.
Hindi family Reels? This is where it actually matters.

What's the wildest health forward your family sent this week?

Reply with the screenshot. I'll run it live.

[Ayan Aggarwal @bot_weiser]

@icanvardar Hi Can, I’d love to help. What kind of design work are you looking for exactly?

[Ayan Aggarwal @bot_weiser]

Huge for the India dev scene.

Cursor + Bangalore + AI builders in one room is exactly the kind of energy we need right now.

See you on 4/25 👋

[Ben Lang @benln]

Cursor team is coming to India:

• Bangalore - 4/25
• Chennai - 5/1
• Mumbai - 5/9

Sign up for the meetups on Luma

[Ayan Aggarwal @bot_weiser]

@xythonix Here’s my portfolio: https://t.co/md3a4poyxl

I focus on clean, minimal mobile product design and have worked on high-scale consumer apps in fintech, mobility, and quick commerce. Happy to share a quick Loom or jump on a call if you’d like to go deeper on process and outcomes.

[Ayan Aggarwal @bot_weiser]

Built a quick AI fact-checking bot this week.

Tested it with:
"The U.S. government has officially declared bankruptcy."

Got back in seconds:
❌ False
+ explanation
+ AP News source

Text-only for now. Core loop works.

#ai #llm

[Ayan Aggarwal @bot_weiser]

Result:

I no longer hit credit limits by 2pm.

Momentum actually stays alive all day. I ship more with the same (or less) budget.

The real unlock wasn’t finding the “best” AI tool. It was finally assigning each tool its proper job.

[Ayan Aggarwal @bot_weiser]

I was burning through AI credits by lunch and shipping slower than I wanted.

Then I stopped treating every AI tool like a Swiss Army knife.

The shift that actually doubled my output:

I stopped using one tool for everything.

Now I run three specialized layers that don’t fight each other:

ChatGPT Desktop → my thinking brain. Strategy, brainstorming, product decisions, all the messy “why are we even building this” conversations.

Claude (terminal) → my pure builder. Heavy lifting on code execution. I keep ChatGPT’s “Work With” feature open so my thinking layer sees exactly what Claude is doing in real time and feeds it better prompts. No more blind handoffs.

Cursor → my repo operator. Refactoring, cleaning architecture, exploring unfamiliar codebases, scoped tasks that agents handle faster than I can.

Zero overlap. Zero context switching between thinking and building.

[Ayan Aggarwal @bot_weiser]

This is gold for staying under Claude’s limits.

But the real unlock for me was stopping the “one tool for everything” game entirely.

Split my stack into three specialised layers → credits last all day, and I actually ship faster.

Full workflow 👇

[kaize @0x_kaize]

https://t.co/ScYWvkqPa4

[Ayan Aggarwal @bot_weiser]

@0x_kaize This is gold for staying under Claude’s limits.

But the real unlock for me was stopping the “one tool for everything” game entirely.

Split my stack into three specialised layers → credits last all day, and I actually ship faster.