How We Built The Legacies.RIP MVP Inside of a Weekend, & What I’ve Learned Along the Way
Six months or so ago, I had the idea to build a platform to make it easier to write meaningful, memorable obituaries while offering support resources to those mourning the loss of a loved one.
After debating whether to embark upon that journey or not, a friend of a close friend passed, and was faced with the mammoth task of trying to memorialize someone whom they loved deeply. Grief-ridden & exhausted, she was overwhelmed by the responsibility. She turned to professional obituary writers, only to find out that the running average rate for a professionally-written obituary was around $450.
That wasn’t okay with me; while there’s a broader conversation to be had about elements of the “death industry” being predatory, as a marketer and amateur technologist myself, I was convinced that recent developments in AI (GPT-3, ChatGPT, generative models & tools) and a little bit of ingenuity could help to relieve this burden.
So with that, I decided to build www.legacies.rip — this is the story of building finding an ally, choosing a “stack”, organizing the project, and sprinting like Hell to get the MVP out the door — all in one weekend.
Part 1: Finding the “Right” Technical Cofounder
-
I’m not technical. I’m tech-adjacent, I can hack my way around some HTML/CSS, I’m a whiz with Markdown, and (as you’ll come to see if you read on) I’m definitely proficient (and certified!) with Zapier, as well as proficient with No-Code tools like Notion, Typedream, Tally, and similar.
But, a developer, I am not. Nor am I a data scientist — amateur or otherwise. Honestly, this has stopped me before from finishing projects… but this time was different: I was *on a mission* and so I swallowed my fears and found the determination to “find a way”.
Simply: I knew I needed help, but I couldn’t afford to hire. Even if I could afford top talent, I couldn’t be sure this would be successful, or how feasible it actually was.
Those proved to be minor bumps in the road, delightfully… in fact, here’s how it panned out.
-
It should first be said that there’s no shortage of services to hire freelance developers — even particular talented ones. In fact, in my career, I’ve actually worked for or consulted to:
- Toptal, where I was employee #6,
- Upstack, where I led Sales & Marketing to great scale & success
- Builder.AI, where I led Enterprise Sales & Marketing
- Codementor/ArcDev, to whom I consulted
- Gun.io, to whom I consulted
— and several others, to boot. All of this is to say: I am certainly aware of the bevy of available options when it comes to retaining expert freelance engineering help for a project.
-
The problems, however, were several:
- I was limited on funds for this project
- I wasn’t sure if it would be successful
- I didn’t even know precisely *what* help I needed
On a whim, I took to *Polywork* and posted about my project & that I was seeking a collaborator to help build an AI-powered empathy-infused professional obituary writer + grief support service for individuals.
Within a day or two, I got extraordinarily lucky; I found Michelangiolo Mazzeschi — an extraordinarily talented AI / Machine Learning expert, data scientist, backend developer, creative, and generally awesome dude.
I still consider myself extraordinarily lucky to have caught Michelangiolo’s interest with my project. So, how do you get a super-senior data science expert to hack on your MVP with you?
Well, I can’t make any promises on that front, but, in speaking with him — and there are some key learnings, here — this is what made the difference with my Polywork post & my approach/pitch to him:
-
- With a ton of AI projects out there, mine stood out as targeting a very specific niche, one that hadn’t been tapped before with GPT-3 tech.
- The distribution/marketing/monetization plan was locked down & solid.
(Legacies is funded by sponsors, affiliates, integration/technology partners, & optional additional custom services — all presented in a non-pushy & assistive fashion. It is, and will forever remain, free for individuals and families to use, fully.)
- Competitive landscape research was done before choosing to move forward.
- A complete Product Requirements Doc (PRD, or “spec doc”) was thoughtfully & methodically written up by me in Linear (here’s a great template that I used from Kavir Kaycee — )
- Willingness — keen-ness, even — to adopt Michelangiolo’s preferred communication & project management stack (a combo of Discord + www.linear.app — both of which I have experience with, fortunately! Thanks Karri Saarinen for building a brilliant product — I hope to work with you, someday!)
- Prioritizing getting paperwork to him detailing a fairly generous revenue sharing scheme to make sure that his efforts were compensated, pending the success of our collaboration.
And last — but ever-so far from least:
- Strong collaborative chemistry, from our first interactions, through our intensive sprint, to our pending launch, and hopefully beyond. It sucks to work with people you don’t like — that’s fortunately been a historical rarity for me, but… if you know, you know.
Part 2: The Setup
-
ADD-brain has been the death of many of my passion projects. I’m not proud to admit it, but, my “stick-to-it-ness” has definitely been a weak point in past efforts. I knew this time had to be different: I was on a mission, and also had someone else depending on me… someone who had already been extraordinarily generous with their time & expertise. I had to get setup properly.
To get the project rolling properly,
1. I started by using PandaDoc (for e-signature as well as template/document editor) to create the Revenue Sharing Agreement & NDA, and getting them over to my new cofounder, *before* starting the work together.
Learnings:
-
- Use templates and e-signature tools to get the paperwork out of the way up front; this keeps everyone on the same page while demonstrating that you have respect for your collaborator(s).
- Be fair — and ideally, generous — with how you split a revenue share / profit sharing agreement… it’s not just about the money: you want the people responsible for your core architecture to be fired up about the success of what you’re building.
2. I propped up a Linear instance, integrated the Discord bot for Linear, and went on an issue-creation rampage.
For anyone who hasn’t heard of Linear, I really cannot recommend it highly enough. It’s remarkably intuitive, I find myself using it to track life tasks along with project work, now. Specifically if you’re fond of tools like Superhuman or Raycast (as I am — *swoon*) — you’ll feel right at home.
Also of note: Linear offers an incredibly generous “Free” tier, replete with unlimited users, 250 issues, & the complete range of integrations/connections… while I’d say they’re more than worth the upgrade spend, you’ll likely find the Free tier to be more than adequate for a side project.
(That said, we’ve crested 60 issues in pretty short order, so, I bet an upgrade is in my near future!)
Learnings:
-
- Organizing your project with granular tasks that are properly dated, prioritized, assigned, & described will foster asynchronous progress between yourself and your allies on a project… “measure twice, cut one”, as it were — spend the time upfront to write out the plan, but pull in your team to advise/revise along the way.
- Don’t make the mistake of letting things play out such that you’re the ONLY one writing Issues (Linear-speak for “tasks”, kinda) in your project tracker… making sure that others know they can write and assign tasks to you will encourage accountability and collaboration.
3. As a matter of priority, I asked Michelangiolo to review the PRD I’d written, and add any potential features / revisions that he might want to — again, using the Linear document editor. (Which is quite nice — Markdown Extended syntax & all; it gobbled up my Notion copy+paste flawlessly. PS — no, I don’t work for/am not an affiliate of Linear — just really please with them.)
-
Some stuff we hacked off the PRD because it was too much for an MVP… other things, which I thought would only be achievable in a v1.5 or even a v2, I was astonished to learn were totally feasible.
Learnings:
-
- Bring the experts into your product plan early. That way, in addition to them having an awareness of what your expectations and desires are, they’ll be able to contribute their own ideas.
- Further, getting on the same page upfront will allow for the architecture to be built from the ground up to support your longer-term goals.
- Don’t limit yourself in terms of ideas! You may be amazed by what is possible… dream big, and bounce the moonshots off of your tech team.
4. We met & screen-shared at 10AM my time the next morning, as a priority. Not only was this a great opportunity to chit-chat and bond a little, but by screen-sharing and walking through things, we were able to get right to work, immediately.
Learnings:
-
- For all of our remote work tools and “hacks”, there’s really no replacement for a face-to-face videochat. I’m saying that as someone that’s pretty video-phobic, and always has been. If you’re asking collaborators to take the jump with you, you’d better be available to them to connect.
With that, we divvied up the work & got to hacking on it! Next up, I’ll share the product functionality we were aiming for in our MVP, how we got there, and some bumps along the way!
Part 3: Rubber Meets the Road
-
Truth be told, I was a little nervous.
I’d lucked out with this brilliant collaborator, I’d written out a killer plan… call it caution, call it imposter syndrome, call it whatever you like: things were going a little *too* well… I figured the “other shoe would drop” once we got to building. Mostly, I didn’t want to let anyone down — including myself. Would I be able to contribute adequately? Was this a good idea? Was it feasible?
The answers to each of those questions turned out to be a *resounding* “yes” — there was nothing we couldn’t handle, in the end, with some creativity and a bit of hackery!
Pretty sure there’s a lesson for *me* in there, somewhere… 🤔
Here’s what we wanted the MVP to do:
-
- Collect some user inputs via some easy/clean/intuitive “frontend” (we were open to — and ended up using! — a no-code solution, so, I’m wary to call it a “frontend”, lest the React Mafia comes for me!)
- Send those to GPT-3 with some clever prompt engineering.
- Get the output back in the form of text
- Format that text into a pretty, well-designed HTML email
- Send that email to the user who entered their information
- Simultaneously send the user’s information (both their written answers & their uploaded photo of their recently-passed loved one) to a dynamic/generative image platform to create custom memorial imagery
- Host that imagery and email it to the user
Sounds easy, right? Doing this in a seamless, performant, consistent, and accessible way was trickier than you might think.
Here’s the full “stack” we ended up using:
-
- AWS Lambda — for handling the data transmission and connection to GPT-3
- OpenAI’s GPT-3 — to take the user inputs & some clever prompt engineering to produce a full written obituary
- Tally.so — for a sleek embeddable form that would talk to an IPaaS provider and pass the data along to Lambda; also, used to collect prospective partner/affiliate/sponsor signups.
- Abyssale — a fantastic platform for dynamic image generation… usually for stuff like ad banners or social media, we used it for portraits of passed loved ones + some of the obit text that came out of GPT-3
- SendGrid — my preferred transactional email provider. We tried Mandrill, Mailgun, MailerSend, even AWS SES — all proved inferior to Twilio’s excellent SendGrid service in terms of getting setup and rocking quickly; no disrespect to the other contenders, half of whom I’ve used extensively!
- DragIt.io — a nice li’l drag x drop HTML email builder that I found from Pitchground.
- Airtable — since the generated & cleaned text came from Lambda, but the uploaded image came from Tally, I used Airtable as an intermediary database of sorts to collate the necessary data for the portrait banner to be generated by Abyssale
- Apollo.io — for finding contacts & writing outreach sequences to commercial client / sponsor prospects.
- Cal.com — for scheduling demo meetings with commercial client / sponsor / integration partner prospects. (SO much love for cal.com!)
- Typedream — for the entire landing page, website, wrapper for the Tally form, and partner registration form embeds
- PorkBun — for domain registration as well as webmail service. (I’m usually a hardcore Google Domains guy, but, the domain was $3.61 on PorkBun & $30+ on Google Domains… not massive savings by any means, but with limited working capital and the stack above tallying up to be $$$, I’ll take it!
- Zapier — sweet, precious, spendy Zapier — how I love thee so. We tried to use Make.com first but it was just turning out to be less than helpful when it came to Tally inputs, sadly. (Huge love for Make, since the earliest Integromat days… but, I’d consider Wade Foster to be a relative friend, too, and am cert’d up with Zapier, so, they win every time.)
So — that’s exactly a dozen tools in total to build this bad boy out.
BTW — if you want links to any of the above tools, or alternatives to them, just drop me a line in the comments, or shoot me an email (address at the end of the article) — I just didn’t want to overwhelm with twelve links in one section.
-
The data-flow looked a little like this:
We worked across two sprints, several hours hours each, Saturday & Sunday, screen-sharing on calls several times & doing lots of testing. The parts that ended up taking the most time were:
-
- Building out the Typedream site
- Writing out the outreach sequences for commercial client prospects
- Fine-tuning (prompt engineering) to get ideal output
- Troubleshooting AWS Lambda shenanigans
- Dealing with DNS & domain verification stuff
- Testing, testing, and more testing.
Learnings:
-
- As a *non* technical cofounder: take advantage of the times that your tech team is building out the platform to take care of as many ops/marketing/growth tasks as you can. There’s ALWAYS — *always* — something to do.
- Don’t be afraid to swap out components in the stack to pursue the best outcomes possible. Finding the right tools for the job might take a bit of time… there are so many options, and so few reasons to settle for less than what you want.
- Test after each change / each new variable in your data flows — lest you have a breakdown in process at the finish line and then have to “play detective” & work backwards.
- Don’t be afraid to hop on the line and screen-share — especially if it’ll be faster than logging an issue / taking a screenshot / explaining (perhaps more than once) — INNTW? (If Not Now, Then When?)
Part 4: Conclusion & Launch!
-
Honestly, building this out was a blast. We’re still fine-tuning things, so, technically we won’t be launching for probably another day or two… we need to do more prompt-tuning, fix some email deliverability issues, and more… but we’re in the final stretch, and I couldn’t be happier/more hopeful.
So, what’s next?
Well, we still have much fine tuning and improvement to do, but we hope to launch within the next day or two.
This week, I’ll be working on:
-
- Prospecting and onboarding commercial partners for Legacies
- Writing out the Grief Support sequence of emails, with guest content & content sourced from the web to help make the grieving process a little easier
- Signing up for affiliate programs to offer our users/subscribers — but Michelangiolo and I are aligned — ONLY services we’ve vetted, NEVER in a pushy way, and EXCLUSIVELY if the service offers referred persons some meaningful/substantial savings.
- Teeing up our launches on ProductHunt, BetaList, and similar
- Doing a little baby PR blitz, just for fun
I suspect Michelangiolo will be working on:
-
- Some hardcore fine-tuning and prompt engineering
- More backend/structural work
- Bringing more of the process into Lambda for the sake of ease
- Experimenting with the Abyssale piece of the puzzle
- (Hopefully!) doing a demo or two to prospective integration partners, with me
At a macro-level, we’re considering bringing on proper frontend help to transition out of no-code tools and into a legit frontend user interface for the platform, but we suspect what we’ve built should serve us well for the purposes of validating the idea & getting an MVP out the door.
Overall, I’ve learned a *ton* during this process, and am proud to have gotten this out the door inside of 48 hours, thanks to a wonderful collaborator and a bit of digital elbow-grease. There’s much more to do, and may be more bumps along the way, but I hope sincerely that this article might be useful to some readers out there — if it encourages just one person to take the leap, today, instead of in six months (or never) — I’ll consider it a massive success.
-
Want to check out Legacies.RIP? Keep in mind that we’re still tuning things up and need just a bit longer to be properly ready.
The link below DOES let one access the platform, which you’re welcome to play with, but for an error-free / higher-variability experience, *please* consider signing up for our waitlist so we can alert you when we go live this week!
—
Grey G. Seymour
Cofounder, Legacies.RIP
