dvny (@dvnyy) header
dvny (@dvnyy) profile picture

20 | ex onchain trader now building neural interfaces

sydney
Joined October 2021
Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)Photo by dvny (@dvnyy)
oooh here's my new website american psycho business card moment so tuff dvnyy.com
3
4
338
Lmk if you have any feedback on it
39
Neurohardware/life day 31 Officially a month in and I feel like I've made more progress than in the past year. Crazy how fast you can move once a goal is visualised. Finished my computational auditory model of the cochlea in MATLAB. Each of the 512 channels are tuned to a centre frequency on a log place map f_p(x) = 7500 * 10^(-0.667*x) to cover 35Hz - 7.5kHz. After two spatial differentiations I modelled the inner-hair-cell by half-wave rectifying the signal followed by a low pass filter. To estimate pitch I computed a harmonic product spectrum for each channel envelope every 20ms, where the largest product is taken as the pitch for that frame. To test pitch perception without the fundamental freq, I designed linear-phase FIR band-pass filters using the Hamming-windowed sinc method, so we can rectify the band-limited signal and find the fundamental freq through (n+1)*f_0 - n*f_0 = f_0 (simulating speech through a telephone). Finally I validated the system against FFT spectra, manual time-domain period measurements, and synthetic harmonic signals and measured a systematic +10% frequency bias and traced it to each section's notch, placed at 0.985*f_p. Quite a fun mini-project that proved to me that something that looks daunting can always be understood and made intuitive by building up from strong foundations. I plan to make a website soon so people can know what I'm working on more clearly.
2
329
I generally agree as long as you don’t correlate struggle with productivity. You don’t have to dread the gym to feel disciplined, and you don’t have to chose the harder option for the sake of it feeling more difficult. You can still challenge yourself while enjoying the process.
A fundamental belief I hold is that you should attempt to make your life as challenging as possible. A lot of people get this completely backwards. When they think about happiness, they confuse it with comfort. This leads to the common false belief that the lack of difficulty is what creates happiness The reality, however, could not be further from the truth. An overly comfortable life is an unhappy life. It is the joy of jumping over the hurdles that brings you satisfaction, not the lack of the difficulties themselves The modern world has been designed in a way to minimize challenges for people. Despite having access to all the resources we could want, people report feeling unhappier than ever before The cause for societal unhappiness is simply due to the fact that we are not pushing our lives to be challenging enough
5
327
Neurohardware/life day 30 Met up with my close friend who dropped out to work at a yc backed startup at the beginning of this year so I didn't do too much work today. It was interesting to see how differently he used ai as a GTM. He used cmux to watch an orchestrator agent break one clear objective into tasks and hand then to subagents working in parallel. He believes that to really get ahead you don't need to be technical anymore, but rather know how to very clearly define tasks that need to be done for the agents to execute. But I also think being deeply knowledgable in your field allows you to elevate the quality of your prompts at a level where entirely blackboxxing it can't. On another note, it's been very nice to see a lot of Sydney based accounts follow me. Please don't be afraid to hit me up for a coffee chat, would love to connect.
1
1
393
Neurohardware/life day 29 For context, I am building a low-power neural recording system that can reliably extract brain signals for decoding. I got 1hr of brain recording from the @neuralink compression challenge. To make this useful I built a Python-based analysis pipeline that separates recordings into different frequency bands (to split the local field potentials (low freq activity) and spikes (high freq)) which we then fourier transform and estimate the power spectral density. Interestingly, I found that 83.3% of recorded power was below 300Hz (in the LFP band) so we could remove low freq activity using analogue filtering before the ADC to save ADC input range for the neural spikes. Also even when no neuron is firing a detectable spike there is small noise which we can estimate using Median Absolute Deviation. It just estimates the spread of background signal without including the spikes that have way larger voltage. Using sd_noise = median*(x - median(x))/0.6745, we get 7.01 LSB. Using this to compare the recording noise against ADC quantisation noise, sd_q = 1/sqrt(12) = 0.289 LSB; thus, sd_noise/sd_q = 24.3, so recording's background noise was 24.3 times greater than the ideal quantisation noise. This suggested that the ADC might be over-specified for this application which I can test by discarding some bits later. Tomorrow I will try to detect potential neural spikes using threshold detection. Truly feels like that constant tormenting inner monologue has finally silenced after a year and a bit of not knowing what I wanted to do. Feels like the university content I aimlessly grinded everyday finally has meaning as bits and pieces of content from random topics show up in this project. I literally can't sleep from how much fun this is.
1
1
1
3
407
I was talking to my quant friend about his feelings on ai. It's wild that there is still bitterness about how fast ai is progressing amongst young stem students. Chatgpt came out in Q4 2022, Opus was released in 2024, and layers like harnesses emerged in 2026. Tell me how you feel disheartened that the most pivotal technology came out during our most progressive, curious-driven years, and that we have an egalitarian fighting chance to directly compete with the monopolies in an otherwise nihilistic late stage capitalist game we would have gotten brutally smurfed in.
1
1
7
305
Wallahi last time I change my username
1
292
Hardware/life day 26 - 28 Began my new project which is a low-power neural recording chain. > Spike detection and lossless compression are benchmarked on Neuralink, Nason and NoMAD data and MEArec ground truth, then implemented as RTL for an ECP5. I used to think the saying 'clarity comes from action' was always obvious but presumptuous since it's hard to take action when you don't know where to begin. I felt mentally tormented trying to figure out what exactly I wanted to do for the past year and it's only when I started building my drone and cochlea simulator that I found newfound clarity in what I truly feel passionate about; neurotech. As long as you have a general idea about what you want to do (electrical engineering + frontier tech), the game is to work intensely to develop skills that can aid you in becoming proficient at said interest and eventually you'll find specific areas that you find fascinating.
1
1
276
If I want to build cybernetics in the future what do I focus on
1
291
Used to struggle to hit 7k steps a day, now averaging over 15k. Some daily habits that helped me increase my steps without affecting my schedule: > walking 10-15min after every meal (blunts insulin spike) > whenever I felt unmotivated (no music or phone) > walk if I wanted to watch youtube/podcast > walking in between sets at the gym > parking slightly further when I drive somewhere > walking in circles at my job lol
walking is so underrated 10k steps a day is a must, bare minimum
1
4
428
Automated most of my dad's business workflow in one night. What used to take him 1-2hrs every night after working from 11-8pm has been freed up. He's asking me for help in other areas as well. People forget how behind the curve a vast majority of the world still is.
1
8
519
Hardware/life day 24-25 Need some advice. Going to korea then backpacking thailand from end of dec - feb so building a new hardware project doesn't sound practical. Shipping the drone fc is going to take a while so we're moving on. Few options: > Build an ambitious project in a field I'm interested in like neurotech and figure out logistics along the way > Make the project software-centric > Focus on research overseas The pragmatic choice is to pick an ambitious project and build until I go overseas and then develop in-depth knowledge through research. Only issue is that I tend to lose motivation if I can't visualise a path to victory and taking a detour by learning without implementing might not be all that optimal for my journey. Although I did thoroughly enjoy studying the theory of my cochlea signal processing simulator (not sure if I'd do that for a month lol) A project allows me to be flexible with the building/learning process like how I used newly yc funded hardware dev tools to build then provide feedback + learn from the founders. Seems like a kill two birds with one stone strategy since I do want to join one of these startups whenever the opportunity arises. What do you think, and any neurotech accounts I should follow?
4
5
492
The retarted nature of crypto sentiment lagging volatile price action is why I felt an incessant need to leave (or at the very least find somewhere else to allocate the majority of my mindshare). Wasn’t there a robinhood season and those bnb and base coins like two months ago? CT had me feeling like it was time to quit my job. I feel like if you’re still in crypto you should either: > be making a shit ton of money or at least have and know you still can or > learn/build something as if it’s your only form of income Your job as a trader is to aggressively cut losses and hold winners, not the other way around.
crypto is so fucking cooked that even the people still here wouldn’t recommend it to someone they care about we used to take insane risks for life-changing returns.. now we’re risking our entire portfolios just to underperform traditional markets while insiders dump on us ask yourself if you’d tell your friends or family to buy crypto right now if the answer is no, why the fuck would anyone new come?
3
369
dvny retweeted
“explain this 6 year gap in your resume” yeah i thought i could make it in crypto
222
829
136
8,216
1,029,929
What people who've experienced success overlook is that your ability to take risk is a muscle that atrophies. Despite your unyielding tenacity and many past triumphs, if you aren't constantly testing new waters, the life you've worked so hard for will inevitably regress back to its former state.
The quality of your life is in direct correlation to your tolerance to uncertainty
5
434
Hardware/life day 15 - 23 Designed the flight controller from scratch and used @diodeinc from @davideasnaghi to clean up mistakes. Built on an STM32H743 and started from a 38x38mm 4-layer to a 50x50 6-layer since the feature set for learned control needed more room. Tested the wiring with Simantic by running the flight software on a simulated version of my board and all the sensors were detected. Saw a post on Diode on my feed earlier today so I messaged the founder and got $50 of credits which I used to fix several issues (before running out midway). I was really interested in how they accounted for sourcing as some of my components had no stock, and replaced them with in-stock alternatives. Had midterms so didn't post but still no excuse for being this lethargic. My current goal is to be more consistent with my projects, keep using new hardware dev tools and immerse more in hardware startups since I'm interested in joining one, and posting more.
1
1
2
358