Milana somi on Farcaster

Content pfp

0 reply

20 recasts

20 reactions

Colin Charles pfp

It’s no longer if you’ll get scammed. It’s when. I’m building Mitsukeru, an always-on AI butler that protects against scams, bad contracts, and social engineering. 💸 $1.03T lost to scams in 2024. ⚡ AI is accelerating infiltration. 🛡️ We need proactive defences, not reactive training. Why here? Because this is grassroots Farcaster Capital Markets, an experiment for the 37 of us here. Raise: $350K Min ticket: $1K Beta ships in 90 days (macOS first) $20/month SaaS model - monetised from day one Your membership = - Utility token allocation at TGE - NFT access pass + perks - Governance + credits Open-source, privacy-first, auditable. Think antivirus, but for scams. Participation is simple. Check the slides, and information in the public pitch. Feel free to DM me here or reply to this thread! Slide deck: https://drive.google.com/file/d/1ql2dFSpnllL9xAUNuN76xeuuuZ4nOKaW/view?usp=sharing Public pitch: https://docs.google.com/document/d/1kW5K56kbUczBYilTR2naqqcueH-rF2Hr0isUdngFFtA/edit?usp=sharing Initial technical roadmap details: https://docs.google.com/document/d/178UDWGGReRyawXeYkOt56BMwmRcooX4Ti09G7pXUaAM/edit?usp=sharing

40 replies

86 recasts

267 reactions

Milana somi pfp

Curious how you’ll handle false positives

1 reply

0 recast

0 reaction

Colin Charles pfp

you ask a great question. because false positives would suck. heuristics evaluation basically, so we do continuous rule refinement with threshold tuning, mix it with machine learning (ML), behavioural analysis, know what a good baseline is, always allow exception lists, and a little more. also, you might know that we could use bayesian classifiers, and weighted scores. a lot of adaptive learning. note i did not once even mention the use of an LLM here, yet ;-) oh, and continuous shipped updates to the host, for fastest responses

1 reply

0 recast

1 reaction