Stories, Money, and Generatively Pre-trained Models; Reflecting on another year of startup journey
?? Amir Feizpour
Helping _you_ use LLMs | Founder @ ai.science | AI Advisor | Speaker | Recovering Quantum Physicist / Data Scientist | Let's Meet -> Virtual Coffee Chat (link below)
I have been writing about our startup journey for more than a year now! I initially set out to do this as a way of celebrating what we achieve every month, but over time it became my way of reflecting and documenting our startup journey!
ps. I'm writing this update a little later than usual to combine my November and December updates in one.
In this update you will read:
Ask:
I usually include a request for help in these updates, but this time my ask is this: I dare you to spend at least two consecutive days doing something that you truly enjoy with no interuptions!
Previously at Aggregate Intellect...
In?October '22?...
See also:?August '22 | September '22
Here is what happened in November and December 2022:
Community Round
We wrapped up our community round that brought our total investment to date to ~$350k USD (from community members and angels). This was our last fundraising campaign for 2022 and other than the monetary aspect, it really pushed us to think about and take concrete actions to make our product more clearly defined and described. This was particularly tough because we had to rethink how to explain a technical product targeted at machine learning developers to people who are less familiar with ML workflows, but also it was tough to explain the complexity of a deep tech product like this without being too generic and hand wavy. I think we made very significant strides in making our story simpler and more concrete through this exercise and I'm excited to continue on this path as we do our first major product launch in 2023.
Creative Destruction Lab Accelerator - Session 2
In the meantime we had to prepare and present at the CDL session 2. What I have started to like about this process is that we have to work on manageable objectives, report on our progress towards them, and get feedback; repeat! What makes this process more constructive (much more, actually) is that between the sessions we get continuous feedback from those who volunteer to mentor us. The mentorship is based on what our objectives are and what mentors think they can help us with. So, overall it's designed to push us in a direction of more focus and creating momentum.
The feedback that we got about our proposed objectives wasn't unexpected, largely echoing what we thought we should be doing next. However, some of the details changed and hearing the mentors' reactions was helpful in determining some details. Overall, I'm still pretty excited about it!
ps. If you have talked to me, you know that I have a cynical view of accelerators because I don't think they're generally set up to help founders, so the fact that I remain excited about this experience says a lot.
The Pitch Story
One of the very early pieces of feedback we got was that there are very nuanced discontinuities and inconsistencies in our pitch (aka growth) story. These are things that we thought we were explaining well, but private discussions with our mentor shed a new light on. I went back to read Liran's old post about this which starts by saying
If you never raised funding before, you might think that the most challenging and important part of the process is getting meetings with angels and investors... After completely tanking those meetings, I realized ... [t]elling an exciting and compelling story is [the hard part]
Yup! been there; done that! Ironically, I had read this blog post back when I started, but reading it again was a completely different experience! Some might say the experience that I have gone through over the past couple of years makes it so much easier for me to see what Liran means here. Importantly it had nuggets for me that I hadn't appreciated the first time but all of sudden inspired me to think about our story completely differently. Of course, talking with him through some of these decision points was also absolutely crucial.
Long story short, I ended up deconstructing our story completely, and rebuilding it with a new lens. This is now the story that I'm most excited about and know exactly what numbers we need to tell an exciting story. Stay tuned for more updates as I get more feedback.
Design Sprint for the Last Mile of our Product
Now, the story of a product can't be told without concrete details about how people use it, can it?
A big chunk of our work in November was making a decision about what use case we wanted to focus on given that we could take our product in a few directions. We started from 12 potential use cases, and narrowed down to 2 after around 70 user / customer conversations. And with some more diligence ended up merging those 2 into one simplified story map. In order to make sure we have thought this through, we ran a week-long design sprint (well, Ammar and Sara did - the rest of us participated). The outcome? a prototype that is fine tuned for our first use case and is ready for some rapid iteration to finalize in conversations with potential customers.
Refactoring our Backend
In more and more of our customer conversations we are hearing that they are excited to try our system but want to make sure their data is private and protected. I've been hesitant to go for that because our system's multi-tenancy is built up in house and I wanted to be way more sure than I am before having people use our platform for sensitive projects. The in house system works for the community side of our platform because everything is communal and shared, but the private access aspects needed work. As part of the AWS Activate program and Nividia Inception, we started talking to AWS architects about the proper ways to ensure multi-tenancy, and tenant data separation, etc. This was also timely because we did need to remodel our data to be more performant as we add more and more data, so we had to revisit a lot of our decisions about how our databases were set up. The good news is that most of our decisions and experiments for the above are well underway and we should be starting the implementation soon.
领英推荐
Generatively Pretrained Models
Well, ChatGPT is hot, eh? (watch our DRT video about it here)
Earlier this year, on a nice summer evening, Percy and I met with Suhas at Queens Park where we discussed some of the new trends in NLP and what we needed to pay attention to. Since then I have been thinking about the emergent multi-task nature of Large Language Models (often generatively pretrained - aka trained on predicting next tokens, but capable of carrying out other tasks like classifying text etc) and how we can leverage them in our product. Our R&D team at McGill and Metropolitan started earlier in the fall and has been investigating ways to improve our current pipeline, and we are increasingly convinced that LLMs are the right way to go. I do believe that the hype around ChatGPT is largely due to very smart product decisions they made and the system they created around an impressive model, and that gives me validation that we are on the right track to go after a vertical GPT-like system that can learn a domain very well and help workers find answers to their technical questions very quickly.
What were the highlights of the year for the team?
Percy: My proudest achievement this year is the backend implementation of the mindmap. As well as starting the research projects with new members of the R&D team. I'm also proud of contributing to refactoring the backend's new data schema.
Next year, I'm excited to have more breakthroughs on the research side with the team. At the same time, I'm looking forward to the new and more efficient graph data schema to power the latest version of our product.
Bereket: My proudest achievement this year is the project mind map feature.
For next year am excited about the backend refactor. It will boost my confidence and I will learn a lot in the process.
Eyob: My proudest moment last year has to be my work on the new library. I get to design and play with different implementation of filter which was educational and better looking.
What I am excited about for the next year is going to be working on rebuilding our backend and learning from that process.
Samuel: Most proud achievement for me this year is stepping up and being there for the dev team when it was needed the most and filling the gap between the development and operation. I have also lead the discussion about changes needed on the backend, and have been in charge of redesigning it.
I am excited about making and managing our system to be simple,?scalable and easy to setup and understand for development so that anyone who will join us can start contributing instantly.
Ammar: My proudest achievement this year would have to be building systems that impacted our culture. For eg, doing fire side style 6-weekly All Hands & starting 6-weekly focus review meets that gave the team a better chance to see how everything fit together & accelerated direction alignment conversations. I'm also proud of activities like the "innovation seeders" [our power users] monthly feedback meetup & a culture of aggressive (near daily) qualitative 'user testing'. This has helped us ensure this renewed focus was directed towards avenues users / customers genuinely cared about. The design sprint we hosted near year end was a culmination of all the above build ups. And the team did beautiful work there to make pivotal user tested breakthroughs.
Proud to have facilitated these changes, and excited to keep the momentum to build a product users love!
Sara: My most proud achievement this year is working with such a good team and learning from them, expanding my knowledge in the field of UX and helping the team to make a product that helps the customers' needs to be met.
I'm most excited about launching the new version of the graph and looking forward to seeing that it helps people in their work.
Parting thoughts...
This has been a transformative year for us, and for me personally. We are at a different place than we were a year ago and have made so much progress in understanding our users' needs and what it takes to tackle them. We could not have been here today without contributions, and even sacrifices that every single team member has made. I'm humbled by all the support we have seen from our community this year, in funding the project and getting their feedback as we iterate on our product. I'm grateful for all the advice and guidance that I have received from advisors and mentors.
I am excited about finally launching our commercial product next year; a journey that has been long but has never been closer!
Resources
Get Started with AI.SCIENCE
Follow?this RECIPE?to know more about what you can do on our platform
Deep Random Talks on YouTube
Want to talk about any of the above? startups? building products? or about your career in ML?
I do (free) 20 minute calls with people to talk about everything?ML?or?startup; if you know anyone who is trying to get into ML / DS, or figuring out how to build an AI product, or thinking about getting into startup life, AND wants to chat about it with me, please pass this link to them:?https://calendly.com/amirfzpr/20min-ama
Mathematician / CEO at Metafold 3D / Solving the toughest geometry problems to unlock manufacturing innovation
2 年Congratulations on all that you and the team have achieved so far Amir! Can't wait to follow your story in 2023!!