
sr. ai engineer | @heyoz_ai ($4m raised) | ex-google soc • ex-linux foundation
Joined November 2021
- Tweets380
- Following143
- Followers158
- Likes3.2K
Voice agents often cut Urdu speakers off mid-sentence, or wait long after they finish.
I built baari, an open Urdu turn detector: 77% balanced accuracy on real Urdu calls, vs 73% for the best existing model.
12 ms on CPU, one-line swap in @pipecat_ai
github.com/samadpls/baari
2
3
56
Just finished a System Design Masterclass.
I got better at estimating traffic, storage, and cost before picking a design.
That’s what I’ll take back to work: figuring out where we’ll hit a limit first, what’s worth scaling now, and what can wait.
#systemdesign
1
3
77
2/
how:
> LoRA with @UnslothAI , 4-bit, ~0.5% of weights trained
> baseline: same models without fine-tuning
> an LLM judge graded each answer vs the reply that solved the thread
Qwen3-8B wrong answers: 31% to 15%
model: huggingface.co/samadpls/torc…
code: github.com/samadpls/torchlig…
1
1
52
released dataset .. torchlight-threads.. PyTorch forum questions, each paired with the answer that actually solved it.
6,841 questions from 22k threads.. Next step is training a model on it.
Dataset: huggingface.co/datasets/sama…
Code: github.com/samadpls/torchlig…
@huggingface @PyTorch
3
5
108
animate-pls now has 20 designs: cutout, clay, comic, linocut, watercolor, terminal and more. Plus reels and edits of your screen recordings.
Your agent writes the video in code, on your machine. It can even narrate in your own voice.
Even this video.
github.com/samadpls/animate-…
1
3
115
After months of work, we're excited to finally launch MedPls! ❤️
An idea that started at a hackathon is now a real app helping families manage medicines, reminders, and prescriptions more easily.
📱 Try it out: medpls.online
We'd love your feedback!
#AppLaunch #Ai
I always wanted a portfolio... but kept putting it off. since I'm not a designer.
I asked @ChatGPT Codex to plan UI, Claude did the mechanical work, and I gave feedback. My first portfolio, and I love it.
The kid in the stamp (me) follows your cursor.
samadpls.com
4
98
When I launched Ask Pakistan, it could search 52 government websites. Today I added 140 more 🎉
It now searches 192 official sites, with 11,849 pages and 1,879 PDFs behind every answer.
Finding the right official page for something simple usually means ten open tabs and a broken link or two. That's the problem I'm trying to fix.
Some of what's new:
• Courts: Sindh, Peshawar and Balochistan High Courts, Federal Shariat Court, Wafaqi Mohtasib, Federal Tax Ombudsman
• Police & safety: Islamabad, Karachi, Balochistan, AJK and GB Police, Motorway Police, Punjab Safe Cities, Rescue 1122, NDMA and PDMAs
• Education & health: AIOU, NAVTTC, TEVTA, PMDC, DRAP, NIH, school education departments
• Utilities: Pakistan Post, WASA Lahore, KWSC, WAPDA, NEPRA, OGRA, PTA
• Business & tax: Pakistan Single Window, SMEDA, Board of Investment, Sindh Revenue Board, KP and Punjab Revenue Authorities
• Overseas Pakistanis: Ministry of Foreign Affairs, Ministry of Overseas Pakistanis
Ask the way you'd ask a friend, in English, Urdu or Roman Urdu 💬 Every answer still links to the official page it came from, so you can double-check it yourself.
It's still an independent project I'm building on my own. It isn't run by or connected to the government. I'm just trying to make official information easier to find 🇵🇰
Give it a try and tell me what's missing 🙌
one interesting design choice in whatsapp is that delivered messages aren't kept forever on whatsapp's servers.
once delivered, your phone becomes the main place that chat history lives.
if you enable backups, those can go to your google account on android or icloud on iphone.
that means whatsapp doesn't have to behave like permanent cloud storage for every delivered conversation.
at whatsapp scale, that's a meaningful architectural difference.
billions of users generate an enormous amount of text, photos, videos and voice notes.
keeping all of that permanently on whatsapp-owned infrastructure would be a very different storage problem.
sometimes an important backend decision isn't how to store more data.
it's deciding what data your backend doesn't need to own forever.
2
60
have you ever noticed cloud providers are very generous when you send data in, but start charging when you send data out?
that's ingress vs egress.
ingress is usually cheap or free because providers want it to be easy to move workloads into their cloud.
egress is different.
once data leaves their network, they have to pay for bandwidth, transit, peering, edge capacity, and the infrastructure carrying that traffic to users or another cloud.
but there is also a business reason.
free ingress makes migration in easy.
paid egress makes migration out expensive.
so over time your data develops gravity.
the more data you store in one cloud, the more expensive it becomes to move away.
this is why network cost matters more than most new developers expect.
compute is not always what gets you.
sometimes it's moving the bytes.
1
32
have you noticed when you switch a google meet from your laptop to your phone, the conversation barely breaks?
that tiny ux detail hides a nice systems idea: make-before-break.
instead of:
disconnect laptop → connect phone
the new device first joins the same meeting, establishes its media connection, and only then does the old device leave.
so for a short moment, both paths can exist.
once the new one is ready, meet switches over.
the general idea is simple:
don't destroy the old connection before the new one is ready.
sometimes seamless ux is just good distributed-systems design hiding underneath.
2
50
a common mistake i see newer developers make is calculating api latency using the average.
and here's the part you might not know: in production, we usually don't optimize latency around the average.
imagine 100 requests come in.
99 finish in 50ms.
1 takes 5 seconds.
the average is still around 100ms.
so your dashboard might say:
average latency = 100ms
sounds fine.
but that number doesn't tell you how many requests actually met your performance target.
this is why we use percentiles.
if we say:
p99 < 100ms
we're saying 99% of requests should finish within 100ms.
we accept that the remaining 1% can be slower because real systems will always have outliers: network delays, cold caches, slow database queries, gc pauses, external apis, etc.
if we need a stricter target:
p99.9 < 100ms
now 99.9% of requests need to stay below 100ms and only 0.1% can fall outside it.
that's the important difference.
average tells you the average of what happened.
p99 tells you whether almost all of your requests are actually meeting the latency target.
next i'll talk about latency vs request time vs throughput because those three also get mixed up a lot.
Content Generated using @HeyOz_AI
2
46











