lmeyerov

72 ms·

751.
by lmeyerov 4y ago
Conversely, I think this is one of the best ways to come up with startup ideas. Learn intimately why dirty work is the way it is, figure out sw to solve it internally, and once that works pretty well... replicate externally. RE:Folks discus
752.
by lmeyerov 4y ago
Awesome, we were just discussing internally how to do such an architecture Maybe a "where to help" / roadmap section for a sense of 'what is left for a stable & reasonably complete 1.0 api'?
753.
by lmeyerov 4y ago
Next step: browser JS VMs need to allow you access gigs of RAM in-memory, it's surprisingly locked down
754.
by lmeyerov 4y ago
The article starts to edge on the switching costs issue and just misses it. Dislodging a great incumbent -- even if the greatness comes from its network effects -- generally takes either the ability to eat it from the inside when marginally
755.
by lmeyerov 4y ago
I see a lot of companies successful via a long-running relationship with MSSPs for improving help desk style support and occasional emergency forensics etc. That's been getting more interesting, and as we partner with MSSP ourselves (g
756.
by lmeyerov 4y ago
Super promising, thanks. Will definitely watch for production readiness - this direction is a big deal as we find most data is cold (logs, docs, ..), not hot, and finally starts becoming more relevant than just doing ES or our own!
757.
by lmeyerov 4y ago
we find mamba solves deps solving for conda (we fail to do GPU dependencies without it), and I think it's getting integrated my main thing w/ conda is it's bananas figuring out how to make a new recipe, which is pretty surpri
758.
by lmeyerov 4y ago
can vespa index 100M+ vectors on a regular RAM cpu server? any faster w/ a gpu (T4 / A10)?
759.
by lmeyerov 4y ago
can this do something like a 100M+ index on a single node? it seems like all the vc-funded oss options are targeting more like 1M-rows-per-server, which doesn't really make sense for most of our use cases..
760.
by lmeyerov 4y ago
Maybe useful: As part of improving autoML in PyGraphistry, we have been a doing a lot of auto-feature-engineering and auto-clustering work for one-liners like: g.nodes(pd.read_csv(...)).featurize().umap().plot() It incorporates a variety o
761.
by lmeyerov 4y ago
if the company keeps growing revenue ahead of hiring , they'll care more about retaining good people than losing them so they can keep growing. in this scenario, hiring a build engineer to fix CI becomes a reasonable thing to do: th
762.
by lmeyerov 4y ago
The good news is you can totally fix a broken CI in a working company, product, & team, so those are the more important things to ask about
763.
by lmeyerov 4y ago
Free drinks and fast CI don't solve a broken company, product, & team. The former you can always fix on your own, but not the latter Two+ years of your life have much more important questions than staging servers, such as: * "
764.
Import PyGraphistry: 100X faster Python imports with lazy loading and Tuna(graphistry.com)
3 points by lmeyerov 4y ago|0 comments
765.
by lmeyerov 4y ago
Graphistry / US DOJ | Data - Multiple | Full-time | US DC/Virgnia metro area (hybrid on-site + WFH) | $150-200K salary; + more in total compensation Graphistry's visual graph AI platform gets used by analysts all the way from
766.
by lmeyerov 4y ago
As a tweak: Follow a top CEO/CTO as person 2-10 with a market you like... ... And then: * Learn how 0->1 , product market fit, and overall process works, which you miss by joining growth-stage * Grow from working with customers hand
767.
by lmeyerov 4y ago
We do webgl rendering of big graphs as part of Graphistry, and before the 20M nodes break the system, the ~10X more edges will ;+) But more seriously, a careful webgl implementation gets you maybe halfway there, and most I've seen brea
768.
by lmeyerov 4y ago
Different perspective, more about individual enablement vs company risk: - Share info that enables independent decision making and actions so your team can help recruit new teammates, sell to customers, choose designs, and at a personal lev
769.
by lmeyerov 4y ago
We are paying users with a growing annual account, this is not about free. We want 1 product for chat, enjoy that Slack solved much of community side of chat (cross-company, ...), and Slack's UX is slightly better than Discord's
770.
by lmeyerov 4y ago
We have a paid slack for our team & customers, and a free slack to support our community, and Slack continuing to kneecap our community zone keeps pushing us towards dropping slack in general. Just as Slack emerged to fill a social netw
771.
by lmeyerov 4y ago
I'm using "dumb" to refer to computationally limited reasoning for even small programs. Think linear regression vs neural networks, or in this case, ML-family type systems (with all the modern trimmings) vs an SMT solver. The
772.
by lmeyerov 4y ago
One project I'm particularly thinking of here was around the weakness & stupidity of the type system, and major issues around scaling (federated code evolution), for what was intended to be carefully typed ecosystem code. The JS as
773.
by lmeyerov 4y ago
Something I still struggle to understand with these systems is why indexing 1-10M things is hard / their sweet spot, when that is basically a small dataframe in pandas or a small SQLite, vs the 10B+ rowa // GBs+ I was expecti
774.
by lmeyerov 4y ago
I much rather blame the type system designers, however heroic and brilliant they may be, than the folks using them. Especially so in scenarios like large & collaborative evolving code bases, legacy, module composition, evolving tools, e
775.
by lmeyerov 4y ago
I wish I had real numbers. So instinct from what I've seen: - a data analyst role rebranded as a DS role will be lower paid than a DE role, maybe 50% diff - an actual DS role is probably higher paid than a DE role, but really depends o
776.
by lmeyerov 4y ago
We and our peers have been busy: - Gov, especially cyber + analytics (we do graph ai + gpu viz stuff) - Supply chain: A slowdown is happening, but COVID, regional conflicts, etc. makes needs for manufacturing etc. companies here super burni
777.
by lmeyerov 4y ago
Sort of. We do a lot on the graph AI + graph viz as powertools behind investigative teams here, and behind the scenes, it's not such a clean story. Some real growers, but most we've seen hit a surprisingly low ceiling: - There jus
778.
by lmeyerov 4y ago
Yes, that was what motivated my comment on for GNN inferencing, or more generally, graph-informed inferencing servers: We generally find it's better to separate DB nodes from pricey GPU parts -- always-on $$$ on db vs during-use GPU (d
779.
by lmeyerov 4y ago
We've been enjoying cudf+cugraph for millions/billions of nodes & edges! For notebook/python users, cugraph is setup well to work with non-graph DB's as well. The blogpost makes sense for graph db users, but the pyda
780.
by lmeyerov 4y ago
It's useful to work backwards from the knowledge a DS needs to be worth their weight. Imagine a small team of $400K/yr DS + $400K/yr DE + ... and whatever hw/sw . So say a $2-3M/yr project driving $3M+ of new growin

More ›