Grover (@GroverInnovate) header
Grover (@GroverInnovate) profile picture

Engineer with an Open Notebook

Roorkee, India
Joined February 2025
Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)Photo by Grover (@GroverInnovate)
It's only a week left before my EPF-7 cohort comes to an end. So, I wanted to share the journey here of what I learned, the things I struggled with, and how it all went. This all started 5 months back with me exploring different project ideas. It was overwhelming at first because I had very little knowledge of the space. I was always amazed by the post-quantum roadmap of Ethereum and wanted to contribute somewhere in that. So, I found my project there: Ream BlackBox Interop Testing for Post Quantam The idea was to extend testing for Lean clients using Hive. At that point, there was a lot I needed to understand before I could figure out what to contribute. In the beginning, I spent time understanding how Hive starts clients, how the tests communicate with them, and how to tell whether a test is actually checking the behaviour we care about. My first merged contribution was a refactor of Hive's validation test setup in July(github.com/ethereum/hive/pul…). It was a small change, removing repeated setup across tests. But it gave me a starting point in a codebase I was still trying to understand. Then I moved on to adding actual validation scenarios: rejecting future blocks, continuing after invalid gossip, and handling the same valid block more than once. That Hive PR got merged in August(github.com/ethereum/hive/pul…). It wasn't just about writing test cases, it involved a lot more things as I had to work through the mock networking setup, block formats, and the runtime needed to run Ream's newer binary. This was one of my first practical lessons: when a test fails, you still have to find out which part failed. It could be the client, your test, or the infrastructure connecting them. I had to understand enough of all three before trusting the result. After that, I started going deeper into the signature side of the project. I was reading about XMSS, exploring signing paths in different clients, and checking existing tests in LeanSpec and Ream. A few things I thought would be useful additions were already covered somewhere else. and other ideas needed infrastructure that wasn't present yet. I kept asking myself whether I was adding something meaningful. Would this test catch a real failure? What would a maintainer learn from it failing? That led to exposing Ream's actual proposal signer through Hive. I wanted the tests to exercise the same signer used during block production. A separate implementation just for testing wouldn't give us that confidence. The Ream PR got merged in September(github.com/ReamLabs/ream/pul…), and I could finally build experiments around that path. One of those experiments found a failure when a signing request reached a slot the key wasn't prepared for, so the request panicked an RPC worker, but a later valid request still succeeded. Even describing that result took care because just saying “Ream crashed” would have overstated what happened. I had evidence of a specific failure, and I needed to be precise about it. I submitted the Hive scenarios(github.com/ethereum/hive/pul…) and a Ream fix(github.com/ReamLabs/ream/pul…). Investigating preparation further also led to a LeanSpec fix with regression tests(github.com/leanEthereum/lean…), where an impossible signing request could keep the service advancing preparation without reaching the requested slot. I also learned to check my tests themselves. I verified signatures independently and deliberately changed inputs to make sure incorrect ones were rejected. A passing result becomes much more useful when you know the test can recognise a failure. Throughout these months, I kept thinking that I should be progressing faster. Sometimes I would spend days reading code or research and end up deciding that I shouldn't implement something. There wasn't always a PR to show for that time. I found that difficult to judge, especially when I wanted to make meaningful contributions within the cohort. But that investigation helped me avoid duplicate work and claims I couldn't support. I started understanding why it was necessary. Then, in October, Ream discontinued its Lean client work. My last PRs remained unreviewed, and I couldn't continue the Ream-focused project as planned. I had more work lined up, so this changed how I thought the last part of EPF would go. I wanted to include this because it's part of the journey too. Projects can change direction while you're working on them. You can put effort into something and still have work left pending for reasons outside your control. Sometimes you have to accept that and figure out what you can carry forward. For me, that's a better understanding of testing, more confidence working through unfamiliar code, and a habit of questioning my own assumptions before calling something a finding. Thanks to Kolby (@Kolby_ML ), Derek (@dsorken1) for their mentorship , and the Ream team (@ReamLabs) and the EF community for the guidance throughout these months. A special thanks to Mario (@TMIYChao) because none of this would have been possible without him. He was my mentor from very first day and I would reach out to him with my problems and he would give me THE BEST ADVICE possible. Truly, a legend. I came into this with very little context and an interest in Ethereum's post-quantum work. There's still a lot I don't understand. But now I have a much better idea of how to take something I don't understand, investigate it, and turn it into a contribution that serves a purpose.
1
1
13
495
GPU Pool is a like a personal project to us; because we genuinely wanted to build from a long time . It’s an interesting tech and it has so many good engineering problems to solve. Looking forward to the future! Thanks @hedera @arkivnetwork for believing in us!
Won 2nd place in the @arkivnetwork and @hedera Track at EthGlobal Mumbai 🏆. We built GPU Pool, a platform for campus students to rent out their laptops, earn incentive and access bigger open source models than what their laptops can run for FREE.
3
2
14
462
Also hats off to an amazing team - @_KxrMa_G @Chainer_Rio @YashSingh166431 The team was complete in itself where different expertises coexisted and it felt like a perfect balance. Had an amazing time @ETHGlobal
1
2
86
Fun day
Day 1 at Devcon 8, in 28 seconds. Thank you for bringing the energy, Mumbai! We'll see you tomorrow.
52
Grover retweeted
Binance Attack is still going on btw there was a fourth sweep of around $48M and the attacker has a fresh wallet now :)
KEY TIMELINES for Binance HACK attack is still ongoing
8
21
3
162
15,386
The kind of content, I absolutely adore
whenever i have to cool some liquid (e.g. coffee or tea) with a spoon i remember the everyday physics stackexchange post where they find the best way to do this. very cool physics.stackexchange.com/qu…
2
66
BlocSoc IIT Roorkee x India Blockchain Week 2026 We're partnering with @IBWofficial, and IBW is sponsoring student passes for our students. 📍 1–2 November · Fairmont, Mumbai 🧵
4
7
25
1,204
Plus if you want to learn more about this journey, I am presenting at @EFDevcon on 4 Nov at 4:20 PM in Workshop 2.
It's only a week left before my EPF-7 cohort comes to an end. So, I wanted to share the journey here of what I learned, the things I struggled with, and how it all went. This all started 5 months back with me exploring different project ideas. It was overwhelming at first because I had very little knowledge of the space. I was always amazed by the post-quantum roadmap of Ethereum and wanted to contribute somewhere in that. So, I found my project there: Ream BlackBox Interop Testing for Post Quantam The idea was to extend testing for Lean clients using Hive. At that point, there was a lot I needed to understand before I could figure out what to contribute. In the beginning, I spent time understanding how Hive starts clients, how the tests communicate with them, and how to tell whether a test is actually checking the behaviour we care about. My first merged contribution was a refactor of Hive's validation test setup in July(github.com/ethereum/hive/pul…). It was a small change, removing repeated setup across tests. But it gave me a starting point in a codebase I was still trying to understand. Then I moved on to adding actual validation scenarios: rejecting future blocks, continuing after invalid gossip, and handling the same valid block more than once. That Hive PR got merged in August(github.com/ethereum/hive/pul…). It wasn't just about writing test cases, it involved a lot more things as I had to work through the mock networking setup, block formats, and the runtime needed to run Ream's newer binary. This was one of my first practical lessons: when a test fails, you still have to find out which part failed. It could be the client, your test, or the infrastructure connecting them. I had to understand enough of all three before trusting the result. After that, I started going deeper into the signature side of the project. I was reading about XMSS, exploring signing paths in different clients, and checking existing tests in LeanSpec and Ream. A few things I thought would be useful additions were already covered somewhere else. and other ideas needed infrastructure that wasn't present yet. I kept asking myself whether I was adding something meaningful. Would this test catch a real failure? What would a maintainer learn from it failing? That led to exposing Ream's actual proposal signer through Hive. I wanted the tests to exercise the same signer used during block production. A separate implementation just for testing wouldn't give us that confidence. The Ream PR got merged in September(github.com/ReamLabs/ream/pul…), and I could finally build experiments around that path. One of those experiments found a failure when a signing request reached a slot the key wasn't prepared for, so the request panicked an RPC worker, but a later valid request still succeeded. Even describing that result took care because just saying “Ream crashed” would have overstated what happened. I had evidence of a specific failure, and I needed to be precise about it. I submitted the Hive scenarios(github.com/ethereum/hive/pul…) and a Ream fix(github.com/ReamLabs/ream/pul…). Investigating preparation further also led to a LeanSpec fix with regression tests(github.com/leanEthereum/lean…), where an impossible signing request could keep the service advancing preparation without reaching the requested slot. I also learned to check my tests themselves. I verified signatures independently and deliberately changed inputs to make sure incorrect ones were rejected. A passing result becomes much more useful when you know the test can recognise a failure. Throughout these months, I kept thinking that I should be progressing faster. Sometimes I would spend days reading code or research and end up deciding that I shouldn't implement something. There wasn't always a PR to show for that time. I found that difficult to judge, especially when I wanted to make meaningful contributions within the cohort. But that investigation helped me avoid duplicate work and claims I couldn't support. I started understanding why it was necessary. Then, in October, Ream discontinued its Lean client work. My last PRs remained unreviewed, and I couldn't continue the Ream-focused project as planned. I had more work lined up, so this changed how I thought the last part of EPF would go. I wanted to include this because it's part of the journey too. Projects can change direction while you're working on them. You can put effort into something and still have work left pending for reasons outside your control. Sometimes you have to accept that and figure out what you can carry forward. For me, that's a better understanding of testing, more confidence working through unfamiliar code, and a habit of questioning my own assumptions before calling something a finding. Thanks to Kolby (@Kolby_ML ), Derek (@dsorken1) for their mentorship , and the Ream team (@ReamLabs) and the EF community for the guidance throughout these months. A special thanks to Mario (@TMIYChao) because none of this would have been possible without him. He was my mentor from very first day and I would reach out to him with my problems and he would give me THE BEST ADVICE possible. Truly, a legend. I came into this with very little context and an interest in Ethereum's post-quantum work. There's still a lot I don't understand. But now I have a much better idea of how to take something I don't understand, investigate it, and turn it into a contribution that serves a purpose.
1
10
375
Reminder : Only a month left before GTA VI
2
58
Grover retweeted
we even got flying cars before gta 6
8
4
1
60
3,426
Youtube music is way too goated
1
40
Still trading on polymarket events? We made it more interesting. Introducing perpetuals on binary markets. Eros lets you trade YES/NO tokens with leverage 😱 Built on @monad Try it out ↓ erosmarkets.vercel.app/ Currently live on testnet TRADE WHAT HAPPENS NEXT. @monad_dev
5
2
26
1,117
Check out this technical blog that explains the design decisions in our project. @monad
1
43
I'm going to ETHGlobal Mumbai @ETHGlobal! Use my referral link below to earn a discount on your stake! Link: ethglob.al/XCDUUS6W58 ethglob.al/XCDUUS6W58
2
64
Building a trading bot shouldn't take a VPS, three API keys and 400 lines of wallet code. In Metropilot, you snap blocks together: → When ETH crosses $3,200, sell 25% → Every day at 09:00, buy 50 USDC of MON That's the whole bot. Built on @monad
Most traders know when they'd buy or sell. They just aren't watching when it happens. Meet Metropilot: no-code trading agents on @monad. Snap a strategy together, rehearse it on paper, then let it run with hard limits on every trade. Your strategy, on autopilot. @monad_dev
2
5
86
Building something really cool, check out!
Most traders know when they'd buy or sell. They just aren't watching when it happens. Meet Metropilot: no-code trading agents on @monad. Snap a strategy together, rehearse it on paper, then let it run with hard limits on every trade. Your strategy, on autopilot. @monad_dev
3
46