Sriram Subramanian
@sriramsubram
Followers
2K
Following
2K
Statuses
2K
Founder and CEO at @niledatabase. Previously: VP of Engineering at Confluent. Believe to win.
Joined September 2012
🚀I am excited to reveal Nile to the world finally! Nile is serverless Postgres database for modern SaaS Nile virtualizes tenants into the database to provide 1. Tenant isolation and per-tenant backups 2. Drop-in organization and user management 3. Global tenant placement 4. Tenant-aware vector embeddings 5. Instant internal customer dashboards 6 Effortless elasticity
34
59
430
The future belongs to "Builders". Creativity, ambition, product thinking, hard work, packaging, and storytelling will determine success
2
0
5
RT @karpathy: I don't have too too much to add on top of this earlier post on V3 and I think it applies to R1 too (which is the more recent…
0
2K
0
The biggest moat for model providers was cost. Lowering the cost to train (and infer) will democratize the model layer. We will see more companies training models, more experimentation and lot more adoption. This is good news for Nvidia and datacenter buildouts. Bad news for model providers in the short term (OpenAI et al) till they provide significant value up the stack.
0
0
2
This is going to only accelerate adoption. We are just getting started.
0
1
2
We ran a 1000+ node sql server cluster at Microsoft to store the metadata for our object store that managed petabytes of data 12 years back. Simple mapping between the object store and the metadata layer helped scale many services including data balancer, data consistency checkers, replication, deletions, compliance rules, data restoration etc. At scale, it becomes really critical to keep the mental model simple. Even the simplest of systems get complex at scale. There were definitely cases of orphaned metadata or blobs but with redundancy, frequent auditing and reconciliation, we were able to build a battle hardened system that ran for more than a decade with high availability.
Concrete example of the difference between theory and practice in systems engineering: When we were designing the storage system at @Dropbox (called "magic pocket") every new PhD graduate would say it was dumb to use a thousand-node MySQL cluster to store the mapping from file chunks to their locations in a million-node storage cluster. Instead they'd point out that a distributed hash table is a much more "efficient" way of storing this mapping with no need for a database. This is fine in theory but in practice you need to 1. run verifier jobs that walk over your indexes and make sure all the data is in the right place 2. dynamically adjust the amount of data on each node or cluster due to hardware issues or data center migrations 3. run queries over the metadata for auditing and forecasting and all manner of things 4. have a really simple codepath for committing a write to the system and not much is simpler than writing a single row to a DB. It turns out the simpler approach is better because the simpler approach works and can be maintained. All this stuff can be learned but it's much much easier to learn in an environment where you run into constraints on a daily basis.
0
15
121
We pre-loaded a bunch of B2B database schemas to make it easy to get started with @niledatabase and Postgres. You can create a DB, select a use case, and have all the schemas loaded in a few seconds.
0
2
5
@criccomini @andy_pavlo yeah, they have to kill one. Same thing with KSQL and Flink acquisition.
1
0
2
Not really. You have a lot of work to get a good multi-tenant system built, which makes it hard to focus on another SKU. It is hard to realize how focus has a compounding effect when companies need to scale. It is much easier to do multiple options at a small scale with few customers. It is also tempting to assume that a single architecture can unify everything - hosted dedicated, hosted multi-tenant, and BYOC. Our goal was this, but we eventually realized they all had 20-30% of architecture that differed a bit, which pushed us to make hard calls. Things could have changed now at CFLT
0
0
1
Lots of misguided info in this thread. - calcium score vs CCTA radiation exposure - what does calcium score of 0 tell you if your age is below 50 - non calcified plaque or soft plaque being main contributor for young folks - benefits of eggs, red meat etc - impact of apob for young folks I have researched this for many years now across many medical professionals. The health knowledge pretty messed up and I still learn and unlearn things even now. I would be careful to advice others about health
Went and got a CCTA heart scan at Cleerly after seeing a bunch of prompting from @altcap. I'm pretty fully bought in on the need to measure and drive your own health.
0
0
2
You can now visualize database schemas within the @niledatabase console ➡️ Easy to manage a large number of tables ➡️Fast and responsive ➡️Manage table schemas and relationship
0
2
10
To build world-class products, you must acquire a strong product sense and know what great looks like. One of the best ways to do this is to use other products often. Not from just competitive products (which might be helpful in some cases) but across all domains. Product taste has become a massive differentiator in modern products and will be even more true in the AI-driven world.
0
0
2
RT @gwenshap: The vision that launched @niledatabase was simple: Databases should do more for developers. DBs should align with applicat…
0
6
0
RT @niledatabase: Happy 2025! Inspired to build a B2B application this year and not sure what to build? Want to try learning AI, RAG or age…
0
2
0
You can create and manage unlimited databases with @niledatabase across all tiers, including the free tier
0
2
6
RT @gwenshap: That time when @v0 explained to me why @niledatabase is an excellent choice for my project:
0
3
0
RT @niledatabase: Sharing our most popular tutorial Build a multi-tenant application with NextJS and Nile's Postgres 👇Doc link https://t.…
0
2
0