Arun ๐Ÿš€ (@arunkc97) header
Arun ๐Ÿš€ (@arunkc97) profile picture

Data Engineer @tenaraiglobal๐Ÿ’ผ โ€ข Opensource | Building schemabrain โค๏ธ

Github ๐Ÿ‘‰
Joined February 2016
Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)Photo by Arun ๐Ÿš€ (@arunkc97)
This is awesome ๐Ÿคฉ๐Ÿคฉ๐Ÿคฉ Our team @bytehash_tech won the runner up in @web3con hackathon ๐Ÿ™Œ๐Ÿ™Œ I'm so happy ๐Ÿคฉ๐Ÿคฉ๐Ÿฅณ๐Ÿฅณ๐Ÿฅณ
gm! we are very pleased to announce @web3con 2022 hackathon winners ๐Ÿฅณ ๐Ÿฅ‡ $15,000 Grand Prize - Brought to you by @Decentology & @0xPolygon: NAT ๐Ÿฅˆ $10,000 Runner Up - Brought to you by @Filecoin, @LivepeerOrg & @radicle: bytehash more winners ๐Ÿ‘‡
10
2
1
25
Arun ๐Ÿš€ retweeted
I just built my own coding harness, just like Claude Code. (100% open-source) A harness is the code wrapped around an LLM. The model only decides the next step, and the harness handles everything else, planning, tools, memory, and safety. In simple words model acts as the brain and the harness provides the hands. A coding harness is that same wrapper pointed at a codebase. It turns a plain text generator into something that reads files, edits code, runs tests, and fixes real bugs. Here's everything that went into mine: - the core agent loop - file tools that double as external memory - planning for long-running tasks - subagents that work in their own context - sandboxed execution in a throwaway VM - human-in-the-loop approval - persistent memory and checkpointing The whole thing is built on CrewAI, a 100% open-source framework. I also wrote an article on this. It covers everything that goes into building a coding harness, the agent loop, planning, subagents, sandboxing, memory, and checkpointing, built step by step. The article is quoted below.
65
237
10
1,273
125,761
How do you stop AI agents from reading or exposing PII? Relying on system prompts is fragile. We built a ๐—ฑ๐—ฎ๐˜๐—ฎ๐—ฏ๐—ฎ๐˜€๐—ฒ ๐—ณ๐—ถ๐—ฟ๐—ฒ๐˜„๐—ฎ๐—น๐—น for AI agents: ๐Ÿ”’ Locked Catastrophic Floor (credentials, cards) ๐Ÿ“„ Config-as-code (git-ready YAML diffs) Try it: github.com/Arun-kc/schemabraโ€ฆ
26
Built this Data Dictionary UI for SchemaBrain with Claude. Schema โ†’ entities, PII, joins, metrics. #BuildInPublic #claude #claudedesign
49
1/ Everyone's takeaway from the 9-second AI database wipe was "just give agents read-only access." Correct. Also, not the finish line. It's the floor.
1
29
6/ Real safety isn't one wall. It's the same rule holding at every layer the agent can reach. Read-only stops the delete. It doesn't stop the leak. Test the layer beneath before someone else does.
1
7
A boundary you ask a model to respect only holds while it feels like respecting it. Turn the temperature up on an agent, and it doesn't just hallucinate โ€” it quietly stops following your "use only these 3 sources" rule too. Enforce limits outside the model, not in the prompt.
1
32
Arun ๐Ÿš€ retweeted
Gmeow Who's excited to play a web3 game that doesn't feel like a chore?๐Ÿ˜ผJust download on March 26th, play, have fun, earn MEOWS with your friends If this is you, then... Gmeow it back!
2,699
20,733
418
38,419
387,824
Dencun. โœ… Memecoin frenzy. โณ The top-two retail-focused L1s are competing for users. But what do users really want? And how do we measure it? โ€ข Speed โ€“ TPS โ€ข Cost โ€“ Avg fee per transaction โ€ข Volume โ€“ Transaction count Most users also want ease of conducting transactions, but that's easier said than measured. Based on these 3 consumer metrics, @Solana is outpacing Ethereum. Not just @ethereum, it's outpacing @base. Solana's: โ†’ TPS is 110x more than Ethereum. โ†’ Transaction count is 9.7x Ethereum's. โ†’ Fee (avg) is lesser than Ethereum and Base. (This is after Dencun and amid the memecoin frenzy though). Will this disparity continue?
2
10
704
If 100 people use a chain, how many will be retained after a month? @osmosis has the highest retention rate at 33% while with @Aptos_Network, only 8 of 100 users remain after a month. Here are the rest: โ€ข @SeiNetwork: 8.8/100 โ€ข @Polygon: 19.4/100 โ€ข @THORChain: 19.8/100 โ€ข @solana: 20.9/100 โ€ข @arbitrum: 25.1/100 โ€ข @Optimism: 27.4/100 โ€ข @avax: 28.6/100 โ€ข @cosmos: 31.7/100
18
20
13
104
44,179
Arun ๐Ÿš€ retweeted
๐ŸŽ Earn Web 3 Giveaway ๐ŸŽ We are giving away 10 FREE passes for @EarnWeb3DAO ๐ŸŽซ To enter: 1๏ธโƒฃ Comment on this post 2๏ธโƒฃ Retweet for an extra chance Winners will be drawn tomorrow! ๐Ÿš€
633
510
33
550
Day 23 of #100DaysOfWeb3 Going through level 9 of the sophomore track @LearnWeb3DAO โšก ๐Ÿ› ๏ธICO -- In Progress๐Ÿคฉ #wagmi ๐Ÿš€ #100DaysOfCode
5
19
Day 22 of #100DaysOfWeb3 Started going through level 9 of the sophomore track @LearnWeb3DAO โšก ๐Ÿ› ๏ธICO ๐Ÿคฉ #wagmi ๐Ÿš€ #100DaysOfCode
1
2
10
Astonishing growth! ๐Ÿคฉ๐Ÿฅณ All the best @LearnWeb3DAO and thanks for all the amazing contents. It's really helpful for all of us Web3 learners ๐Ÿš€๐Ÿš€
WEโ€™RE HIRING! ๐Ÿš€ After 8 months of buildingโ€ฆ we are expanding the team! Send us your Resumรฉ! Say hi ๐Ÿ‘‹๐Ÿ‘‡ angel.co/company/learnweb3-dโ€ฆ
1
9
#Data101 - Byte sized #data info Data Ingestion ๐Ÿ’‰ is the process of absorbing data from a vast multitude of sources, then transferring it to a target site where it can be analyzed and deposited. There are mainly 3 data ingestion strategies used widely
1
3. Upsert load Upsert load is a kind of variation of the incremental load. It's basically update + insert. Upsert is an operation that will update an existing row if a specified value already exists in a table, and insert a new row if the specified value doesn't already exist
1
Picking up what kind of strategy to be used for your project depends on the use case. Every strategy has its own pros and cons. Thanks for the read!! If you found the thread useful, give a follow โœ… likeโค๏ธ, and retweet the first thread โ™ป๏ธ
1