Data Platform at Doordash. Previously at Uber and Netflix Cloud Platform. Interested in Data scale challenges and managing engineering excellence. Hiring now!
Joined April 2008
- Tweets1.8K
- Following470
- Followers1.2K
- Likes1.3K
Sudhir Tonse retweeted
Your teamโs AI bill doubled ๐ซ . Is that wasted spend, or money well spent?
Before you can answer, you need to know where the increase came from. More people using AI? A few heavy users? Unnecessary Opus 5.5 and Astra 6 use?
At Galileo (acquired by @Cisco) and now @splunk agent observability, we were piecing that story together across five AI providers, each with its own definition of usage.
Here's another qs - Whose budget does that usage belong to? Knowing someoneโs manager doesnโt automatically give you the answer. In fact, finance may attribute your manager's spend to a different category than your spend...
Weโve been building tokenomics in Splunk Agent Observability to help employees understand their usage, managers understand adoption, and finance connect spend to budgets.
That gives people a starting point for deciding whether the spend makes sense :).
We learned A LOT building the tokenomics product for CRAZY different ICPs - Engineers, Finance folks and Management ..... all while building it for the scale of a 100K ppl at Cisco - our first customer.
I wrote all about in my blog (in ๐งต), please tell us what you'd wanna see in a product like this.
what does tokenomics mean to you?
Sudhir Tonse retweeted
Excited to announce that @PierreB80788038, creator of ZeroFS is joining @tensorlake to help us build an elastic file storage on top of object store for sandboxes!
This would enable sandboxes to have a shared file system between them.
It was great hanging out with all the meetup attendees and doing a demo of Multi Agent Observability and Governance. Good talks from Google Gemini and Modal as well. Read more at AgentControl.dev
This Post is from an account that no longer exists. Learn more
Well NZ said in the press conference they would aim to silence the crowd.
By that they meant by putting them to sleep!
Love this framing!
When it comes to building Agentic Systems, making it Eval driven surely pays off strong dividends in ensuring you have built it right and can continue to monitor your investment using solid metrics and alerts.
Iย have been developing Agentic Systems for the past few years and the same patterns keep emerging. ๐
๐๐๐ฎ๐น๐๐ฎ๐๐ถ๐ผ๐ป ๐๐ฟ๐ถ๐๐ฒ๐ป ๐๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐บ๐ฒ๐ป๐ is the most reliable way to be successful in building your ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐ฆ๐๐๐๐ฒ๐บ๐ and continue improving them - here is my template.
Letโs zoom in:
๐ญ. Define a problem you want to solve: is GenAI even needed?
๐ฎ. Build a Prototype: figure out if the solution is feasible.
๐ฏ. Define Performance Metrics: you must have output metrics defined for how you will measure success of your application.
๐ฐ. Define Evals: split the above into smaller input metrics that can move the key metrics forward. Decompose them into tasks that could be automated and move the given input metrics. Define Evals for each. Store the Evals in your Observability Platform.
โน๏ธ Steps ๐ญ. - ๐ฐ. are where AI Product Managers can help, but can also be handled by AI Engineers.
๐ฑ. Build a PoC: it can be simple (excel sheet) or more complex (user facing UI). Regardless of what it is, expose it to the users for feedback as soon as possible.
๐ฒ. Instrument your application: gather traces and human feedback and store it in an Observability Platform next to previously stored Evals.
๐ณ. Run Evals on traced data: traces contain inputs and outputs of your application, run evals on top of them.
๐ด. Analyse Failing Evals and negative user feedback: this data is gold as it specifically pinpoints where the Agentic System needs improvement.
๐ต. Use data from the previous step to improve your application - prompt engineer, improve AI system topology, finetune models etc. Make sure that the changes move Evals into the right direction.
๐ญ๐ฌ. Build and expose the improved application to the users.
๐ญ๐ญ. Monitor the application in production: this comes out of the box - you have implemented evaluations and traces for development purposes, they can be reused for monitoring. Configure specific alerting thresholds and enjoy the peace of mind.
โ
๐๐ผ๐ป๐๐ถ๐ป๐๐ผ๐๐ ๐๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐บ๐ฒ๐ป๐ ๐ผ๐ณ ๐๐ผ๐๐ฟ ๐ฎ๐ฝ๐ฝ๐น๐ถ๐ฐ๐ฎ๐๐ถ๐ผ๐ป:
โก๏ธ Run steps ๐ฒ. - ๐ญ๐ฌ. to continuously improve and evolve your application.
โก๏ธ As you build up in complexity, new requirements can be added to the same application, this includes running steps ๐ญ. - ๐ฑ. and attaching the new logic as routes to your Agentic System.
โก๏ธ You start off with a simple Chatbot and add a route that can classify user intent to take action (e.g. add items to a shopping cart).
Join me in a free webinar this Friday to learn how LLMOps patterns fit into this picture: maven.com/p/7d5864/llm-ops-pโฆ
What is your experience in evolving Agentic Systems? Let me know in the comments ๐
Sudhir Tonse retweeted
Iย have been developing Agentic Systems for the past few years and the same patterns keep emerging. ๐
๐๐๐ฎ๐น๐๐ฎ๐๐ถ๐ผ๐ป ๐๐ฟ๐ถ๐๐ฒ๐ป ๐๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐บ๐ฒ๐ป๐ is the most reliable way to be successful in building your ๐๐ด๐ฒ๐ป๐๐ถ๐ฐ ๐ฆ๐๐๐๐ฒ๐บ๐ and continue improving them - here is my template.
Letโs zoom in:
๐ญ. Define a problem you want to solve: is GenAI even needed?
๐ฎ. Build a Prototype: figure out if the solution is feasible.
๐ฏ. Define Performance Metrics: you must have output metrics defined for how you will measure success of your application.
๐ฐ. Define Evals: split the above into smaller input metrics that can move the key metrics forward. Decompose them into tasks that could be automated and move the given input metrics. Define Evals for each. Store the Evals in your Observability Platform.
โน๏ธ Steps ๐ญ. - ๐ฐ. are where AI Product Managers can help, but can also be handled by AI Engineers.
๐ฑ. Build a PoC: it can be simple (excel sheet) or more complex (user facing UI). Regardless of what it is, expose it to the users for feedback as soon as possible.
๐ฒ. Instrument your application: gather traces and human feedback and store it in an Observability Platform next to previously stored Evals.
๐ณ. Run Evals on traced data: traces contain inputs and outputs of your application, run evals on top of them.
๐ด. Analyse Failing Evals and negative user feedback: this data is gold as it specifically pinpoints where the Agentic System needs improvement.
๐ต. Use data from the previous step to improve your application - prompt engineer, improve AI system topology, finetune models etc. Make sure that the changes move Evals into the right direction.
๐ญ๐ฌ. Build and expose the improved application to the users.
๐ญ๐ญ. Monitor the application in production: this comes out of the box - you have implemented evaluations and traces for development purposes, they can be reused for monitoring. Configure specific alerting thresholds and enjoy the peace of mind.
โ
๐๐ผ๐ป๐๐ถ๐ป๐๐ผ๐๐ ๐๐ฒ๐๐ฒ๐น๐ผ๐ฝ๐บ๐ฒ๐ป๐ ๐ผ๐ณ ๐๐ผ๐๐ฟ ๐ฎ๐ฝ๐ฝ๐น๐ถ๐ฐ๐ฎ๐๐ถ๐ผ๐ป:
โก๏ธ Run steps ๐ฒ. - ๐ญ๐ฌ. to continuously improve and evolve your application.
โก๏ธ As you build up in complexity, new requirements can be added to the same application, this includes running steps ๐ญ. - ๐ฑ. and attaching the new logic as routes to your Agentic System.
โก๏ธ You start off with a simple Chatbot and add a route that can classify user intent to take action (e.g. add items to a shopping cart).
Join me in a free webinar this Friday to learn how LLMOps patterns fit into this picture: maven.com/p/7d5864/llm-ops-pโฆ
What is your experience in evolving Agentic Systems? Let me know in the comments ๐
Sudhir Tonse retweeted
If youโre a distributed systems engineer excited about database-adjacent systems, container infrastructure and cluster schedulers, weโre hiring at @tensorlake.
Our team has built systems at HashiCorp, AWS, Netlify, and SurrealDB.
Build foundational AI infrastructure with people who care about correctness and scale.
Our platform lets developers build agents and workflows on a durable, sandboxed, serverless runtime, handling fan-out, retries, and correctness under failure.
Below: Claude Agent SDK powered agentic web crawler running on Tensorlake!
Hey SWEs, PMs and aspirational AI builders, here is a free course/materials to up your game or learn what Evals/Observability are and why they are an important part of your skillset as you embark on your journey of building AI based applications!
linkedin.com/posts/sudhirtonโฆ
Sudhir Tonse retweeted
Over the past few years, DoorDash Labs has been building one of the most sophisticated autonomy stacks designed for the real-world challenges of local delivery.
Today, we're introducing Dot, the first commercial autonomous delivery robot to travel on bike lanes, roads, and sidewalks. At one-tenth the size of a car reaching speeds of up to 20 mph, Dot is purpose-built for local commerce.
Together with our new Autonomous Delivery Platform, Dot is the next step toward a more efficient, sustainable, and accessible delivery ecosystem for everyone.
Sudhir Tonse retweeted
Roughly one third of a modern CEO's job is trying to find that fucking Slack conversation you were in earlier
Sudhir Tonse retweeted
Build Your Own "Git" In C From Scratch
-really great playlist, use this to get most out of this.
-it's great if someone wants to understand how really "Git" works behind the scenes.
-it would give you solid foundation for "Low level system programming".
-don't watch completely, it's quite long, just do what you need.
-learn as you require concepts that you need.
Sudhir Tonse retweeted
10/ When reviewing your products and roadmap, ask:
- What are the key feedback loops?
- Where are the delays in the system?
- What are the hidden incentives?
- Which metrics reinforce current behavior?
- What mental models drive decisions?
Follow @nurijanian for more product thinking from a technical perspective.
I share frameworks and practical guides in prodmgmt.world if you want to dive deeper into product thinking.
Sudhir Tonse retweeted
Fundraising is harder than everybody thinks, even if youโve raised billions before.
Here's fundraising advice I wish I had when I started ๐งต
Sudhir Tonse retweeted
Hash Table is the most important data structure out there and apart from knowing how to use it well, it is even more interesting to know how it is built and the internal details.
Some time back, I was on a spree to dissect the inner workings of HashTables to its core including the math behind it to build a detailed and practical understanding. I compiled all my learnings in an 11-video series
give it a watch - youtube.com/watch?v=jjW8w8EDโฆ
I went through 2 books, 4 papers, 13 articles, and 3 codebases to understand them end to end and if you are a curious engineer and love diving deep, you'll love it. The topics I have covered in the series are
- internal structure of a HashTable
- collision resolution through chaining and open addressing
- collision resolution through linear and quadratic probing
- what determines the performance of the hash table
- why are HashTables always doubled in size when they grow
- how to implement HashSets and HashMaps
โก I keep writing and sharing my practical experience and learnings every day, so if you resonate then follow along. I keep it no fluff.
youtube.com/c/ArpitBhayani
#AsliEngineering #Engineering #HashTableInternals
Sudhir Tonse retweeted
Over the past month or so, both me and @shirshanka been inundated with curious questions on #datalakehouse and #data #catalogs.
We thought it ll be fun to exchange ideas and think out loud across both these areas together.
So tomorrow, Thursday, July 11th - 9am PST. Its on. Weโre going to try something different.
Tune in to the Linkedin Live event
linkedin.com/events/ddl-ep08โฆ
Sudhir Tonse retweeted
We are happy to see more excitement in this space and are working faster than ever to make data open across clouds, data warehouses and compute engines.
Here's a blog focusing on the Apache Hudi lens, taking stock of the current state of the Hudi land and what the future holds.
#opensource #data #datalakehouse #datawarehouse #cloud #apachehudi #dataengineering #analytics
onehouse.ai/blog/apache-hudiโฆ