Fork me on GitHub

So #C05P6UP4G2K just celebrated its 23rd birthday and #C05SVE3P116, its 4th. Three-ish weeks and four whole days. It feels much longer to me. I guess it also depends on how many hours are counted as a day. Lack of sleep isn’t a badge of honor. It’s an unfortunate and untenable characteristic of the situation. The creation of #C05SVE3P116 is really an admission of failure; I have failed to figure out my project quickly. It seems Life isn’t a weekend or few-weeks project, at least not for someone with my current set of skills. I’m bad at this, bad on several levels. And I can’t help feeling “bad” here is good. Reality is the only ground upon which to build. Unlike in golf, we can only play the ball where it lies. My Givens 1. The challenges associated with human development and interaction are worth the effort; more, they are interesting. 2. I am my first client. 3. Initially working with transitions of wealth makes sense given the frequency of failure and above-average risks associated therewith. Wealth transition is the first area of emphasis. 4. When measured against the need ($58T in the US), there are nearly zero effective practitioners of holistic wealth transition facilitation. Not zero; nearly zero. 5. There are several types of businesses that will buy Wealth Transitions (a subset of Life). “Buy” here means both product purchase and enterprise purchase. The initial incentive will be marketing and it will be laced with cynicism. They know firsthand the (heretofore) intractable nature of the issues. When they figure they can both retain and grow inter-generational business, they will throw money. These types of firms regularly argue for ascendency in the wealth transition pecking order: a. Asset management b. Law, particularly those specializing in estate planning c. Accounting d. Business consulting e. Financial planning f. Multi-family office 6. The dearth of effective practitioners and tools is born of systemic complexity and professional turf wars. Professional certifications make interdisciplinary work difficult, sometimes impossible. Regulatory compliance often makes interdisciplinary work illegal. 7. The software and the consultancy have to progress together; they are co-emergent. 8. The work is necessarily a research effort on many levels and in different senses. 9. Natural language (NL) must be is the core UI tech. Other auxiliary interfaces will develop. 10. A panoply of fine-tuning and embedding profiles will be required. 11. Knowledge graphs (KG) are the core guiding structures. KGs keep NL resources coordinated, intentional, and serve as anti-hallucinogenics. 12. The Clojure community and the tools that arise out of it constitute the best environment for the work of which I am aware. With all due respect to Python, I cannot afford to be in a v3-breaks-v2 culture. Python is at best in recovery, though it doesn’t quite know from what. 13. The core Clojure tech is reliable, fast, practical, and is as consistent across the stack as is currently possible. As best I can tell, these attributes be being forwarded generally. 14. AWS is a generally reliable, strange, evolving uberOS. My Questions I’m swimming in questions. I share them here to share them, not expecting engagement on any of them, never mind all of them. Articulation helps. • Do I just keep plodding on building as best I can with all education coming in the form of OJT? • Do I go through a bunch of tutorials? Which and how many? Don’t I need to be more architecturally settled to answer “which and how many?“? Chicken or egg? • Do I build a pre-MVP with Biff to understand Clojure basics (an orchestrated tutorial)? • In which order do I add components to my system? (auth, payment, embedding, LLM “smoothing”, LLM fine-tuning, etc.) Just auth has unusual on-boarding requirements. • Do I study AWS in one way or another or do the OJT there too? • How much exercise does the core model need before it’s useful? • Do I find/hire a coach? • Do I just go find funding and hire an experienced developer? • What parts of my thinking are sound? Which aren’t? How do I distinguish between the two? • Where does the serenity prayer apply (other than “everywhere”)? • Has someone already built a NL interface to a set of KGs using Clojure/Datomic? • Do I find a NL/KG system elsewhere and translate it? Which one? • Where am I overestimating my capacities? Where am I underestimating? • How long until I can start using Life with clients? What kind of deadline(s) do I set? • Am I going to starve before I can be useful? That’s where things are now. Your thoughts are deeply appreciated. What will the 24th and 5th birthdays bring?


Re: Given #12: I went to PyCon in 2013 (it was local and therefore "cheap" to attend and the core team were talking about adoption of v3 and that by about 2018 they expected everyone would have switched to v3... another five years later and v2 is still alive and well. 😞 • OJT -- I think learning by doing is probably the best way. • tutorials -- I can't say; some people learn better that way and just keep plugging away at them. • biff -- This would be my recommendation: Build something you can start to use and leverage the added value of Biff to get you up and running quickly. • priorities -- pick the "most important feature" to get you to the next stage and build that (I'd imagine auth then the LLM stuff so you can let some early testers use it for free while you build out more features, but you are the "Product Owner" so you may have a stronger sense of "most important feature"). • AWS -- OJT again, when you need it. I think Biff favors Docker and Digital Ocean as the first "cloud deployment"? Follow its lead and you can switch you ECR/ECS etc at AWS when you feel you're ready or need that sort of scalability. • core model -- no clue; the #C054XC5JVDZ channel that might have more insight (I'm not a member but I see you are). • coach -- no clue; I suspect you can get part-time ad hoc mentoring from a few folks here but I think you need a plan of what you want to get out of coaching first. • hire a dev -- a possibility; do you want a technical co-founder (paid partly in equity in your "company") or a genuine contracted developer? (the latter will likely be much more expensive but the former requires you to give up some control in the project) • soundness -- I can't speak to your domain; I think just picking a given tech stack and building something (see MVP/Biff above) will move you forward and allow you to verify your ideas more. • serenity -- I suspect a degree of "analysis paralysis" right now so accepting where you are, picking a stack, and moving forward will give you something to control 🙂 • prior art in Clojure -- no clue; ask #C054XC5JVDZ and maybe #C0AMXL6LQ (and #C03RZMDSH I guess)? • translate existing -- I suspect these things exist as Java libraries so just use them from Clojure via interop? • over/under -- building a full-stack NL/KG app with an NLP UI/UX is definitely a challenging task; trying to identify "small bites" (a la elephant) and eating those might give you better insight into what you can achieve yourself and what you really do need help with. • deadlines -- not all clients are the same so find one or two really "sympathetic" ones and treat them as the direct users/testers of your MVP; ask them what's the "most important feature" and try to build that for them. Again "small bites" where possible. • starving -- how are you supporting yourself now? How much runway do you have?


Bottom-of-my-heart thank you, @U04V70XH6. The smartest thing I’ve done is find you people. I just spent 3 hours on a Google Meet with @U0175247L06. Brilliant. • OJT — That has indeed worked best for me — plug away • Biff — I think so too for a real MVP. I’ve been concerned about the effort to convert to industrial scaling. If I’m not making enough money to pay for help, I’ve got bigger issues. • Most Important Feature — The talk with Omar nailed it: I need to walk a Q&A graph on a single page. I don’t need NL resolution yet. • AWS — Yes, it’ll be there when I need it. The emphasis on ASAP product is so helpful. • Core model — To be honest, I know it’s solid. It just needs to be populated. Omar turned me on to Logseq. I’m assuming I can export a graph out of there somehow. • Coaching objective — basic product up and running so I can get to work. • Partner — I’m more inclined to find a tech co-founder. It’s like finding a spouse. Long-term, a partner is much more expensive. The engagement is the what’s valuable (no pun intended). Omar is thinking about it. I’d have to come up with a prospectus of some sort. • Serenity — It’s just a big territory to contemplate. That said, I can see exactly what needs to be built. Speed matters. • Prior Clojure — I asked #datomic. Both #clojurellm and #clojure-nlp are good suggestions. I didn’t know about #clojure-nlp • Existing non-Clojure — Lots of Python. I’ll keep looking. • Bites of elephant — The one-page-interview is the first bite. • Guinea pigs — Finding the right clients is key. • Starving — I’m going through the couch for loose change. I ran out of runway awhile ago. Gotta figure out how to get in the air without tarmac. I’m off to Biff and Logseq (for the model data). Thank you again and again.


> Existing non-Clojure — Lots of Python. There's always and #libpython-clj ?


True that.