Senthil
@besenthil
Followers
262
Following
2K
Media
90
Statuses
2K
Engineering Leader @witnesschain | AI x Blockchain | Building verifiable infra proofs
Bangalore, India
Joined May 2008
There is a reason why garbage trucks are enclosed. They prevent littering, reduce smell and improve overall hygiene. In a lot of countries, the laws mandate and enforce it.
0
0
0
https://t.co/onvpf3pykQ Is this the live dashboard page for me to track potholes in Bangalore?
0
0
0
Why it matters: PoL helps crypto connect to the real world. 🌍 It can verify: - Compute nodes running in specific regions - IoT devices or drones where they claim to be (accuracy differs based on the number of watchtowers around that claimed location) - Local DAO participation
1
0
0
The result: No GPS. No screenshots. No trust. Just math + latency. And once verified, the result goes on-chain. 9/10
1
0
1
Security features: You can’t challenge yourself Only nearby nodes (within 2000 km) can verify Nodes must be active and publicly reachable All responses are cryptographically signed 8/10
1
0
0
Before all this, the system "calibrates" Watchtowers test pings between each other to learn how distance maps to delay in their region. It’s like building a "ping = distance" formula for every area. 7/10
1
0
0
Witness Chain does this globally. Prover -> node claiming a location Challengers (Watchtowers) -> nodes with known, verified coordinates They exchange latency data, sign it cryptographically, and triangulate a verified region. 6/10
1
0
0
"If his reply took 20 ms, he must be within ~5 km of me." Each one draws a circle around themselves - "You’re somewhere inside this circle." When multiple circles overlap - that’s your verified location. 5/10
1
0
0
So imagine this: You (the prover) say: "Mom, I’m really at the office." A bunch of coworkers nearby (the challengers) send you quick pings: "Hey, reply to this!" They measure how long your response takes. 4/10
1
0
0
Here’s the trick: Witness Chain measures how long messages take to travel between devices on the internet. Because, just like sound, the farther you are, the longer it takes. That delay is called network latency. 3/10
1
0
0
But crypto doesn’t trust selfies. Photos can be faked. GPS can be spoofed. So Witness Chain built a way to mathematically prove location - no photos, no location trust. Just cryptography + physics 2/10
1
0
1
Explaining @witnesschain 's Proof of Location (PoL) like I’m explaining it to my mom Mom, you know how you ask me "Where are you?" …and I say "I’m at the office, working late." But you don’t always believe me You say: "Send me a selfie, so I know you’re really there."
1
0
1
EigenCloud mainnet-alpha unlocks something big: 🔸 Verifiable compute (EigenCompute) 🔸 Verifiable AI (EigenAI) 🔸 Proof-of-Personhood (PoP) 🔸 Proof-of-Location (PoL) — powered by @witnesschain Together, they enable the next frontier: verifiable cyberphysical apps. @eigenlayer
now that eigencloud is in mainnet alpha, it’s exciting to explore what crypto developers can build at the intersection of eigencompute, eigenai, proof-of-personhood (PoP) and proof-of-location (PoL). quick recap: > eigencompute: unstoppable + verifiable offchain runtime >
0
0
1
We have shipped PoL, PoB and Image Verification Services. All on top of @eigenlayer trust! @witnesschain
PoL has always fascinated us Great to see @witnesschain & @eigenlayer are bringing it to DePIN & AI Now an agent can verifiably prove its location in the physical world 📍 Kudos to @0xranvir & the team
0
0
2
On Arattai, we have initiated discussions with Sharad Sharma of iSpirt, the group that did the technical work to make UPI happen, to standardize and publish the messaging protocols. I am a huge fan of UPI and hugely respect the work the team did. Sharad is a good friend and he
728
3K
19K
Instead of a human spending hours trying “add step-by-step reasoning” or “use You are a helpful assistant,” GEPA automates that experimentation - using the model itself to reflect, mutate, and evolve prompts. Read more here:
arxiv.org
Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of...
0
1
5
Centralized AI gives us power — but also limits. Models locked behind paywalls, controlled by a few, with little room for community contribution. But what if it didn’t have to be this way? What if you had the power to choose what model to run — directly from a peer in the
0
3
4