building @actuity | building https://xitter.cf/t.co/xe9zux3yGM | sometimes @metamemoryai | ai engineer

New Delhi, India
Joined March 2021
"Until death, all defeat is psychological." the last 3-4 months have been full of failures. more than i can count. now i have around 2.5 months left in 2026, and i'm setting a goal to make $10k before the year ends. i'm open to freelance work, AI agent development, automations, and building technical products from scratch. if you need something built, or know someone who does, my DMs are open. let's see if i can turn these last few months around. but i've still been building. gifsy.fun : turn text or images into proper 3d models. 3,000+ users in the last 1.5 months. actuity.tech : send AI agents to your website to test whether they can actually navigate it, understand it, and complete tasks. 6k+ web visitors in the last 10 days, with 600+ people trying it. alongside this, i'll keep growing both gifsy and actuity while working towards that $10k goal.
4
4
20
997
fuck, I've been literally looking for a good photos app for mac because current one sucks gotta try this one
I got bored of the default Photos app on my Mac, so I decided to build one of my own. Introducing PHOTSY 🖼️ A beautiful home for your photos, screenshots, and videos. Relive your highlights, convert images and videos, and create PDFs, all in one place. No converter websites, no uploads. Photsy is officially live - photsy.app
1
3
312
Aman Jha retweeted
hyperframe gotta be the goatest software ever to exist
2
9
316
hyperframe gotta be the goatest software ever to exist
2
9
316
Context selection for RAG, clearly explained: Your retriever returns 24 chunks. Most pipelines send all of them, or just the top 5. Treat the prompt like a knapsack instead: 1. Score each chunk: relevance × recency × novelty 2. Recompute novelty every round, so a near-duplicate of something you already picked is worth almost nothing 3. Fill the token budget by value per token 4. If one chunk alone beats the whole set, take that instead 5. Don't drop the rest. Squeeze it into a 2-line summary On LongMemEval-S this kept 90.3% of the answer evidence on 79% fewer tokens. Top-k at the same spend kept 78.8%. Open-sourced it as CABAR: npm i @whyparabola/cabar
2
3
232
Aman Jha retweeted
Today, I'm launching Vene: your AI advocate for healthcare. I used to spend a dozen hours a week managing care for my dad, a disabled veteran, and my mom, who has a severe autoimmune disease. Healthcare was a full-time job for my family. Our goal is to end the burden of care for every family: When you're sick, or caring for someone who is, the hard part is everything after the hospital: the referral that never arrives, the test result nobody calls about, the bill that makes no sense, and everything in between. For my dad, he spent hours on the phone to book an MRI. When he showed up, they told him to his face that insurance had stopped covering it. Then insurance told him to call a list of other offices and get the paperwork sent over himself, and he ended up giving up. That’s why we’ve trained Vene to handle the toughest aspects of healthcare: - Coordinating several specialists at once - Challenging surprise bills and treatment denials - Catching missed follow-ups and care gaps My mom uses Vene to coordinate her primary care and endocrinologist, automatically refill and renew her prescriptions, and manage my little sister’s dental appointments. They were my first users, and now over 5,000 families use Vene, and we're deploying it to another 50,000 patients across provider networks in the US.
284
166
245
931
305,305
is market really that cooked or trust in kirat is going down? by now it should have been oversubscribed because kirat is indeed a big name in the Solana community it's been only 14 hours but still and ik it's going to achieve its target, but I assumed it's going to get oversubscribed in few minutes am i missing something?
There's one bet in poker that never loses. I'm making it. Welcome to The Inner Table. The first hand hasn't been dealt yet. This is the only time you can sit down before it is. Raising on chain on @solana @backabledotbiz backable.biz/raises/BGyEf4Ew…
8
24
4,781
avg dau to 27 from 20 in a day, we should start doing marketing aggressively most of the traffic is coming from chatgpt, need to double down on it gifsy.fun
1
1
122
we got Shravani cooking mac app before GTA VI
A little sneak peak!😋 Any guesses??
1
4
166
Aman Jha retweeted
I am happy to share that I have finally able to create a skill I wanted for posting on @Reddit specifically! repo link - github.com/piyushjha0409/red…
2
3
30
810
avg dau rose to 20 from 11 in a day, not bad gifsy.fun
2
105
Aman Jha retweeted
"Until death, all defeat is psychological." the last 3-4 months have been full of failures. more than i can count. now i have around 2.5 months left in 2026, and i'm setting a goal to make $10k before the year ends. i'm open to freelance work, AI agent development, automations, and building technical products from scratch. if you need something built, or know someone who does, my DMs are open. let's see if i can turn these last few months around. but i've still been building. gifsy.fun : turn text or images into proper 3d models. 3,000+ users in the last 1.5 months. actuity.tech : send AI agents to your website to test whether they can actually navigate it, understand it, and complete tasks. 6k+ web visitors in the last 10 days, with 600+ people trying it. alongside this, i'll keep growing both gifsy and actuity while working towards that $10k goal.
4
4
20
997
Claude Code tip: before you pay for a bigger GPU, have Opus 5.5 time every step of a cold start. Gifsy's photo → 3D took 4–5 min. I asked why. It measured: - loading 15 GB of weights: ~100 s - recompiling GPU kernels in every fresh container: ~140 s - the actual model: 13–17 s An H100 saved 3 s warm and was slower cold. The fix: keep the kernel tuning on the volume, restore the loaded model from a GPU memory snapshot. Now ~1–1.5 min from cold.
3
1
7
211
now you can generate 3d model like this using gifsy.fun
1
2
5
363
Aman Jha retweeted
can you sponsor us @vercel or else we're gonna broke😭
1
6
317
Aman Jha retweeted
past 24 hours stats, and most of it is coming from chatgpt DAU is also increasing
1
3
5
206
Aman Jha retweeted
Auto-rigging a 3D character from one photo, clearly explained: 1. Redraw the character in a T-pose first. Arms glued to the body can't be rigged. 2. Turn the redraw into a mesh (TRELLIS), cut to 60k faces. 3. Fit a humanoid skeleton template (UniRig) and give the bones Mixamo names, so any animation fits. 4. Check the rig. Left and right swapped? Rig again with the next seed. 5. Skin it. Legs share weights only at the crotch. A rifle or backpack sticks to one bone instead of bending. 6. Bake idle, walk and run into the GLB as clips. This is what runs behind every character on Gifsy now.
2
2
5
331
can you sponsor us @vercel or else we're gonna broke😭
1
6
317
past 24 hours stats, and most of it is coming from chatgpt DAU is also increasing
1
3
5
206