Adrenaline motivated rustacean nomad. @SuperteamIn, prev: @RomeProtocol, @MangoMarkets, @MLHacks. Fixing agave and its read layer.
Bengaluru, India
- Tweets 1,018
- Following 1,967
- Followers 565
- Likes 13,981
Pinned Tweet
Read layer fixed. Live mainnet replay, 32 GB RAM box, 12 cores:
• 1.13B accounts / ~497 GB state
• 0 slot lag, ~2.5 slots/s — pacing the cluster
• 86% read-cache hit
• 0 bank-hash mismatches (5,300+ slots)
• wiped disk → live in ~20 min
Data freedom for @solana is coming. More soon.
5
25
2,310
Aniket Prajapati retweeted
28 days. 28 states. 1 school.
I'm building a school in a village called Solana, funded by a token on Solana, and documenting the entire journey across India. One state a day.
Introducing $SLATE
CA: EU5WWxWBjDzfs3pvb6N7yuc6gBfo5Ltao6qfdXDiH2U3
solanaschool.in
85
64
366
51,548
Aniket Prajapati retweeted
Few more
Checkout triton GitHub many read layer repos to contribute
github.com/orgs/rpcpool/repo…
Blueshift (spbf, infra, pqc etc..)
github.com/blueshift-gg
4
15
478
These 2 IIT’s killing it. Give @TrenchersAI a shot.
“AI optimized it” isn’t enough.
- show me the old setting
- show me the new one
- show me what, when and why it changed
If the Agent can change the strategy, the trader should be able to audit it.
1
3
6
561
Aniket Prajapati retweeted
1 in 100 product brands run a fleet & they get 27x the views
7
10
19
1,392
Fintech finally discovered tx splitting. Make it all on chain!
App Idea - A UPI QR code for merchants that automatically breaks a single large transactions into multiple smaller ₹2000 transactions.
And the govt cannot count it as tax evasion, because according to them MDR is anyways not a tax right 🤷♂️
1
4
383
Aniket Prajapati retweeted
Very interesting result: frontier model robot performance did not change with thinking effort.
5 effort levels, 2 simulators, 135 episodes, one variable. 34× the reasoning tokens bought exactly nothing.
Astra is an Incredible step up from GPT 5.6 at robotics tasks
- Astra scored 10/10 vs GPT 5.6 Sol at 3/10
- The step budget is a lot higher since frontier llms are slower.
Astra is very strong at fine-grained spatial reasoning and action sequencing.
1
6
619
Aniket Prajapati retweeted
For an adversarial leader it is unless the batch fills the entire block?
they can just enable searchers to sandwich the batch within the block as from what I’ve read there is nothing stopping a custom node from doing so.
With all the slot lagging & validator shenanigans it seems most likely this is where it will end up unless it’s a full block per batch?
2
3
440
21-28th sept. BLR. see ya fellows there !!!
introducing ascent — a DeFi residency for the ones on the frontlines. 21-28 september. BLR.
build from india, for the entire world.
come join the ascent: ascentresidency.com
1
1
7
690
I just hope this doesn't just remain gated to Triton and Helius. It should be accessible via Double Zero or a public tap in.
BAM Preconfirmations are now live via @Helius and @triton_one
Get access to the lowest latency data stream from validators running BAM-compatible clients; covering over 34% of network stake
1
2
333
A lot of indexers are going to go down today on @solana
2
11
804