Venkatesh Seetharam

Lots of interesting talks at @InMobi Encore event on AI and systems. Love to meet builders at the event at lunch or dinner. DM me.

@tech_optimist @DSPyOSS I’ve several questions 1. Support for tool calling? 2. Support for agent loops? 3. Multi Agents are modeled as many modules? 4. How do you track progress across these agents? 5. How does an agent handoff a task to a sub agent? 6. How many examples are good for optimization?

The "six pieces" of @DSPyOSS (imo). Start by defining the LM. Then define a signature (user intent + structured types). Then define the module (program logic). Then define examples to evaluate the system. Optimizers are optional when you want to improve the system. Adapters are there when you need them, but worth knowing about.

@Suhail This is an issue specific to India. I’ve had a few folks who don’t work until night and then can’t get things done. They don’t come upfront and discuss they have multiple jobs.

PSA: there’s a guy named Soham Parekh (in India) who works at 3-4 startups at the same time. He’s been preying on YC companies and more. Beware. I fired this guy in his first week and told him to stop lying / scamming people. He hasn’t stopped a year later. No more excuses.

A wonderful hand curated community of AI builders came together today in #Bangalore to listen to pioneers in the space with good food. Kudos to @pingali and @vikramsrinivasa for putting this together.

@ashpreetbedi Real world applications have complex DAGs and this looks very simple - need retries for steps, decision nodes, etc.

Agents, Teams, and Workflows - 3 patterns that cover the full spectrum of Agentic applications. Agents are precise, Teams are autonomous, and Workflows give you control. Here’s a workflow that caches previous inputs, it’s just Python ✨

@martin_casado Very well said. This is nothing new - think of LookML, DAX, RDF, OWL, etc. this is our bet too.

Re: Programming. We've been moving to a post language world for awhile now with all the libraries, runtime systems, frameworks etc. So syntax has been waning in importance. And AI will of course accelerate that. However, semantics remain important as ever. And natural language don't accurately describe them. As long as people need to describe systems, they'll need formal methods to do so.

@martin_casado Anyone building anything serious should have a lot of guardrails already else it’s fail spectacularly. Looks like spaghetti but patterns will evolve over time.

Dealing with the unpredictability of LLMs in code reminds me a lot of network programming. You end up having to implement retries, back off, and sometimes have to overlay transaction identification for idempotency. I wouldn't be surprised if this emerges as a formal layer.

#RAG is fundamentally limited. In our most recent post, we describe a more flexible approach inspired by declarative query processing in databases. We call this LLM-powered Unstructured Analytics — #LUnA. To learn more check out: aryn.ai/post/rag-is-a-… #Aryn #Sycamore

As we reach mid-January, let us check in on our resolution. Have you been looking for motivation or inspiration to help you with starting a transformational path toward your goals? These are a few exercises to help you make 2024 your year of growth, endurance and realising your personal goals. Read the full article: heartfulnessmagazine.com/ground-yoursel… @heartfulness #newyear #motivational #exercises #transformation #goals

@Mashah08 Can you elaborate what you mean by LLM powered data query engine?

#RAG is fundamentally limited. It assumes answers are in the top results, and puts responsibility on static embeddings to anticipate all questions. #GenAI is remarkable at sifting through unstructured data, but RAG handicaps it. We need #LLM-powered dataflow query engines.

@BGMusings @1ndus @FoundersFunda @shreesha @rammahesh Somehow it’s always the after party that’s very cool and fun!!!

@1ndus @FoundersFunda @shreesha @innerzeal @rammahesh Aah, too bad I missed it! :-) But the unplanned dinner at Zareen's in Palo Alto afterwards with you all was magnificent!

Holiday party with 250+ desi founders in Mountain View 🎄🎉 @FoundersFunda @shreesha @innerzeal @rammahesh

@FoundersFunda community brought together the Bay area entrepreneurs buzzing with enthusiasm and aspirations. Keeps getting better every year. Thanks to our sponsors!!!

#Aryn is out of stealth. We raised a $7.5M seed from #Factory, #8VC, and others. We released a new #conversational #search stack for unstructured enterprise data. It’s powered by #LLMs, #OpenSearch, and #Sycamore, and all 100% Apache-v2. blog.aryn.ai #GenAI #dataprep.

We are incredibly fortunate to join Mehul Shah and Benjamin Sowell on this exhilarating journey with Aryn Lets go! cc: @BGMusings @kevinniechen @8vc

Aryn is out of stealth, unveiling a $7.5mm seed round and groundbreaking conversational search stack to ask meaningful questions of unstructured data. Congrats @Mashah08 & friends! Some thoughts from @BGMusings @innerzeal & @kevinniechen: 8vc.com/resources/the-…

@bernhardsson Splunk Universal Forwarder?

Has anyone implemented log forwarding to third party systems? How does it work? Obviously asking for a friend.

@schrockn I used Google Voice at Hortonworks and my boss wasn't very happy. I use your.cloaked.app that generates phone numbers and forwards the calls preserving privacy. Its very cool I used to ignore calls from unknown numbers until my daughter was lost and I ignored 4 times

If you ever found a company, never, ever use your real phone number for anything ever. I get 5-15 spam/sdr calls a day now.

The very best part of the Snowflake summit was meeting the humble legend Thierry at our reception. Thanks to @BGMusings for putting this together. Had a wonderful conversation around Data Cloud and AI.