Siva Ragavan (@ragavan_com) profile picture

Know your wiring

Joined October 2008
Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)Photo by Siva Ragavan (@ragavan_com)
Siva Ragavan retweeted
Pinecone has introduced Nexus, a knowledge engine built specifically for AI agents. Instead of repeatedly searching documents with traditional RAG, Nexus compiles knowledge once from sources like data warehouses, CRMs, documents, and Slack into structured artifacts that agents can query directly. According to Pinecone, Nexus delivers: - Up to 30× faster responses - 90%+ accuracy - Up to 90% fewer tokens than conventional RAG workflows As AI agents become more autonomous, efficient knowledge retrieval is becoming just as important as the model itself
2
6
29
2,605
Siva Ragavan retweeted
Pinecone Nexus is in Public Preview. A knowledge engine that compiles untapped, distributed enterprise data into a layer agents query directly. Answers get faster, cheaper, and more accurate. Read more: pinecone.io/blog/pinecone-ne…
2
4
1
20
1,891
Siva Ragavan retweeted
Our Founder and Chief Scientist @EdoLiberty just kicked off the Search & Retrieval track @aiDotEngineer. "Agents don't need to be smarter. They need a Knowledge Layer." We've got a big update on this coming tomorrow...
1
4
16
883
Siva Ragavan retweeted
Most #AIAgents don't fail because of the model. They fail because of the infrastructure around it. Introducing the Nebius Agents Blueprint: an open architecture for building, operating, and continuously improving agents in production. nebius.com/blog/posts/introd…
6
28
10
261
27,543
Siva Ragavan retweeted
Proud to have Pinecone Nexus be a part of this along with our friends at @LangChain, @tavilyai, and @guardrails_ai!
1
2
6
798
Siva Ragavan retweeted
Four weeks, three enterprise customers, same pattern across all three. Most inference spend goes to retrieval loops. A generic index carries no knowledge of domain, query types, or task structure, so the loop runs before the model can reason. Nexus compiles before the query. Full results: pinecone.io/blog/nexus-ea-be…
2
2
5
447
Siva Ragavan retweeted
Replying to @Azure
@Azure OneLake as the System of Record + @pinecone Nexus as the System of Knowledge = Enterprise AI in production, at scale.
We just announced the integration of our knowledge engine, Pinecone Nexus, with @Microsoft OneLake at #MSBuild. Want to build a reliable, production-grade knowledge layer directly over your structured data in OneLake or Fabric? Staff Data Engineer Simon Lu shows you how it's done in this quick demo. Watch here and read the full announcement: pinecone.io/newsroom/microso…
1
2
79
Siva Ragavan retweeted
95% token reduction. 30x faster execution. 90%+ task completion. Today at #MSBuild, we announced a major shift to move reasoning upstream: Pinecone Nexus now integrates directly with @Microsoft OneLake. Traditional AI agents waste tokens stitching raw data together at runtime. At scale, completion rates plummet 60% because we are forcing reasoning engines to do the heavy lifting of data infrastructure. Pinecone Nexus x Microsoft OneLake fixes the infrastructure wall, moving reasoning upstream. ❌ Naive approach: Agent queries raw data ➔ Stitches chunks ➔ Dumps massive context window into frontier LLM ➔ High latency + exploding token costs. ✅ Nexus approach: Nexus integrates with OneLake ➔ Pre-assembles task-specific artifacts ➔ Agent queries structured data via KnowQL ➔ Instant, cited responses. The data your agents need already lives in OneLake. Stop rebuilding data pipelines. Enforce RBAC/ABAC instantly, and track tokenomics via a unified dashboard. 📣 Announcement: pinecone.io/newsroom/microso…
3
4
1
12
1,149
Recently @pinecone introduced Nexus – a new knowledge-engine layer for AI agents that reduces token use by up to 90%. It’s built on top of a vector database, but shifts reasoning earlier in the pipeline: from retrieval at query time to knowledge compilation before the agent even asks for information. Instead of giving agents raw files, Nexus prepares task-optimized representations in advance. These representations include artifacts – structured forms of information tailored to each agent’s specialization and workflow. For example, a finance agent gets only billing schedules, pricing rules, and usage thresholds. Pinecone’s VP of Product Jeff Zhu explains it like this: "The artifact representation can range from markdown files to extracted entities to tables. These artifacts are then indexed for retrieval within Pinecone’s database which support both semantic, sparse, and full text search capabilities so that the right artifacts are retrieved during query time and then composed for the final structured output response from Nexus." So vector databases are evolving from passive storage layers into active knowledge engines for agents. Read this to learn what distinguishes vector database in the Agentic Era: turingpost.com/p/agentic-vec…
8
33
3
198
14,645
Siva Ragavan retweeted
“We used to memorize the whole syntax”
52
427
23
6,316
104,943
Siva Ragavan retweeted
Our internal AI agent was a brute-force nightmare. 🛑 It burned 40,000 tokens, took 2 minutes, and only hit 68% accuracy just to answer a single question. Why? Traditional databases are built for humans. Agents don't have that context, so they just guess inside expensive LLM prompts. We realized we couldn't keep pushing raw data into the LLM. We needed to bring the context closer to the data itself. That led us to build Pinecone Nexus, a knowledge engine that compiles the exact context a machine needs before it hits the LLM. 🧠🏗️ When we moved our agent to Nexus: 📉 Tokens: 40k → 2k (90% drop) ⏱️ Latency: 2 mins → <500ms 🎯 Accuracy: 68% → 90%+ Our CEO @ashashutosh sat down with @a16z General Partner Peter Levine to share our customer zero story and how we're rewriting the AI stack. Stop burning tokens on search. Listen here: 👂Spotify: open.spotify.com/episode/2RA… 👂Apple: podcasts.apple.com/us/podcas…
2
7
1,113
Siva Ragavan retweeted
Pinecone Marketplace, live today. Until now, building an internal app that answers questions over a company's own docs, contracts, tickets, and wikis meant months of retrieval pipeline work: embeddings, ranking, citations, evals. Pick a template in Marketplace, connect the sources, publish. Every answer cites back to its source. Templates for customer support, legal search, sales enablement, onboarding, and more. Free on Starter: 1M input tokens/month through June 30 (2x the usual) so you can test it on real workloads. pinecone.io/blog/marketplace…
2
6
1
15
726
Today we shipped Pinecone Nexus, and I couldn't be more excited to finally be able to talk about it! For a long time, when an agent needs to figure out which version of "correct" applies to its task — which policy, which date, which scope, which audience — that work has happened at inference time, with the model doing its best on the fly. Nexus moves that work somewhere it can actually be done right. Knowledge gets compiled into structured, governed contexts before an agent ever touches it. The knowledge curation happens up front, deliberately. There's a real token bill underneath all of this: Every agent call today is paying — in tokens — for the model to do that disambiguation work in real time, reading through partially-relevant content, figuring out which fragment applies, and often retrying when it picks wrong. Do that work once and it stops being a per-call cost. Amazing work by @SivaRagavan, Jeff Zhu and the entire @pinecone team. More awesome stuff coming this week, so stay tuned!
Introducing Pinecone Nexus. A knowledge engine for agents. The bottleneck for production agents isn't the model. It's the per-query work of searching, stitching, parsing tables, normalizing labels. Nexus compiles that work ahead of time into typed, governed artifacts. Agents query through KnowQL. 90% fewer tokens. >90% task completion. 30x faster. Early access open. pinecone.io/blog/introducing…
1
1
1
2
295
Siva Ragavan retweeted
Assembling context is complicated — searching, stitching, parsing, normalizing. Every. Single. Query. Pinecone Nexus moves that work upstream so agents get typed, governed, cited artifacts instead. And the numbers are hard to argue with: 90% fewer tokens, >90% task completion, 30x faster. Read more from @SivaRagavan and Jeff Zhu and get in on the early access action ⤵️
Introducing Pinecone Nexus. A knowledge engine for agents. The bottleneck for production agents isn't the model. It's the per-query work of searching, stitching, parsing tables, normalizing labels. Nexus compiles that work ahead of time into typed, governed artifacts. Agents query through KnowQL. 90% fewer tokens. >90% task completion. 30x faster. Early access open. pinecone.io/blog/introducing…
1
1
4
174
Siva Ragavan retweeted
Could happen 😅
182
1,450
158
16,560
407,021
Siva Ragavan retweeted
"it worked on production just like it did on localhost"
126
1,149
57
14,294
518,333
Siva Ragavan retweeted
This image really tells the whole story. Nexus makes your agents faster, cheaper, and more accurate.
1
2
100
Siva Ragavan retweeted
The agentic era doesn't fit inside the old infrastructure category. So we built a new one. This week we're unveiling the largest release in the history of @pinecone : a new language for agents, a new engine to compile context, a new marketplace, new economics, and new reach.
What happens when the primary users of AI aren't humans, but agents? This week: A new language, new engine, new marketplace, new economics, and new reach. The agentic era needs its own infrastructure category. Our CEO @ashashutosh Founder @EdoLiberty on what it is and what comes next. With perspectives from @Box, @llama_index, @LangChain, @Teradata, @ThoughtFocusTec, and @UnstructuredIO. pinecone.io/blog/knowledge-i…
4
5
279
Siva Ragavan retweeted
Something's coming. Launch week May 4.
3
3
4
13
3,332