About
Venture capitalist focused on early stage enterprise software startups
Articles by Villi
Activity
12K followers
-
Villi Iltchev shared thisThis is the most exciting and powerful data pipeline project I have seen in a while. If you're building coding agents or any type of knowledge base that you want to make available to agents, CocoIndex is an incredibly powerful tool. Check it outVilli Iltchev shared this🚀 CocoIndex V1 is live. It's the incremental context engine we've been rebuilding for the agent era. Long-horizon agents run roughly 50x faster than humans. They write code, record their decisions, and update their own context as they go. Meanwhile, most of the data infrastructure they rely on is still built around a human's morning dashboard. That cadence doesn't hold up when agents are in the loop. By the time the nightly rebuild finishes, the context is already behind. CocoIndex has always had one job: keep derived data in sync with source data, without redoing work that's already been done. V1 is a fundamental redesign, shaped by a year of watching what teams actually build with Coco. Here's what that looks like in practice: 🔹 A live index of your codebase, a knowledge graph from your meeting notes, or a vector store over your docs can stay fresh without you having to think about it. Edit a paragraph and only the affected chunks get re-embedded. Delete a file and its rows go away on their own. Swap the embedding model and the column re-syncs. 🔹 You can stand up a continuously fresh context pipeline in an afternoon instead of planning a quarter of work around it. Point CocoIndex at your source, describe what you want in the target, and it keeps Postgres, VectorDB, Neo4j, Kafka, S3, or even a folder on disk in that shape as things change. 🔹 The pipelines agents actually produce are finally expressible end-to-end: entity resolution across sessions, clustering, multi-phase reduction, per-tenant topologies, conditional targets. If you can write it in Python, you can run it incrementally. 🔹 Multi-tenant setups stop being a separate project. One Postgres schema per tenant, one Kafka topic per category, one knowledge DB table per entity type, all driven by a live registry. When a tenant or category goes away, CocoIndex cleans up after it. 🔹 Going from `pip install` to a running pipeline takes minutes. The engine is embedded, so there's nothing to stand up alongside it. You prototype locally and ship the same code to production. The mental model is still the clearest way to describe what CocoIndex does. You declare what the *target* should look like as a function of the *source*, and the engine works out the transitions. Think React, but for data. Building a production-worthy incremental engine in-house usually takes 10 to 20 engineers and at least six months, and then keeps asking for maintenance forever as sources, targets, and schemas evolve. CocoIndex absorbs that work into the engine, so what you write is the pipeline itself, not the plumbing around it. Huge thanks to the 70+ contributors who have shaped CocoIndex since the v0 days, and to everyone who lived on the 50 alpha releases with us on the road to V1. - Read the announcement: https://lnkd.in/gi3M3hrV - Star on GitHub: https://lnkd.in/ezksH98y #AI #AIAgents #DataEngineering #OpenSource #RAG #Python #EnterpriseAI #AIInfrastructure
-
Villi Iltchev reposted thisVilli Iltchev reposted thisMore than excited to announce Uncork Capital's investment in Soma Energy, alongside my good friends, Villi Iltchev at Category Ventures and Semil Shah at Haystack. There is increasing and well-founded concern that the world will not be able to produce enough electricity to support the expected insatiable demand of AI data centers over the next decade (we're expected to build 5x the current number of data centers in the next ten years). Athanasios Caramanolis, the Founder & CEO of Soma, and his team is building an important part of the solution. I share some thoughts about why we're so excited to partner with Ath. Uncork's view on seed stage investing is straightforward: we back exceptional teams, building cutting edge products, for big ass markets. Soma checks all three. A world-class team with direct, hard-earned experience, leveraging modern AI to tackle a market constraint that is both urgent and inevitable. https://lnkd.in/g2tyAR3PInvesting in Soma Energy: Making Every Megawatt Count - Uncork CapitalInvesting in Soma Energy: Making Every Megawatt Count - Uncork Capital
-
Villi Iltchev reposted thisVilli Iltchev reposted thisPodcasts are one of the richest sources of expert knowledge on the internet. A single Lex Fridman or Dwarkesh Patel episode can contain dozens of substantive claims about people, technologies, and organizations — but it's all locked inside hours of audio. You can't query any of it. You can't cross-reference what two different guests said about the same topic. In this post, we'll build a CocoIndex pipeline that turns YouTube podcast episodes into a queryable knowledge graph. The pipeline downloads audio, transcribes with speaker diarization, uses an LLM to extract structured statements and entities, resolves duplicates across episodes, and stores everything in SurrealDB as a graph. We use CocoIndex and Pydantic to build the pipeline. CocoIndex is a data indexing framework for building incremental data transformation pipelines — it tracks what's been processed, so re-running the pipeline only processes new or changed episodes. CocoIndex makes it exceptionally easy to build a knowledge graph without writing Cypher, and easy to add/remove any podcast incrementally. The full source code is available at CocoIndex Examples -- https://lnkd.in/g9QgHeh8 ⭐ Star the repo if it is helpful! #AI #LLM #CocoIndex #Podcast #MultiModal #EnterpriseAI #DataInfrastructure
-
Villi Iltchev reposted thisThrilled to be supporting Antioch on their mission to build the simulation platform to bring testing and validation for robotics and autonomy from the physical world into software. Incredible team!Villi Iltchev reposted thisWe’ve raised $8.5M in seed funding, valuing Antioch at $60M, to build the simulation platform for physical autonomy. The raise was led by Category Ventures and A*, with participation from MaC Venture Capital, BoxGroup, Icehouse Ventures, Abstract, and angels. Antioch brings robotics and autonomy development entirely into simulation. Teams build, iterate, and test their full autonomous stacks in a single platform, accelerated by agents with native understanding of 3D environments. We're doing for physical autonomy what coding agents have done for software development. When the entire development loop lives in simulation, engineers and agents can reason about a robotic stack the same way they reason about a codebase: run it, observe the result, and iterate. The funding will go toward further expanding the engineering team, the core simulation infrastructure, and our new agentic framework. TechCrunch covered the raise and what we’re building: https://lnkd.in/g_UFyD8K If you’re an exceptional engineer interested in redefining autonomy development, we’d love to chat: antioch.com/careers.
-
Villi Iltchev reposted thisExcited to share that we've raised $8.5M in seed funding for Antioch, led by A* and Category Ventures, with participation from MaC Venture Capital, Abstract, BoxGroup, Icehouse Ventures, and a group of angels we're lucky to have behind us. A year ago, Harry, Colton, Collin, Michael, and I started Antioch with a conviction that simulation would become the foundation of how autonomous systems get built and tested. We'd all seen the same thing from different vantage points: teams building fake warehouses and running expensive field trials, spending months calibrating simulated sensors and physics to match real-world behavior, and wrestling with tools like Isaac Sim and Gazebo that were built for researchers, not production engineering teams. We're using the raise to double down on what we think the future of this space actually looks like: autoresearch for physical autonomy. Coding agents work because the feedback loop is tight: write code, run it, observe the result, iterate. Physical autonomy hasn't had that loop because you can't spin up a warehouse on every commit. Simulation gives you exactly that runtime, and we're building the agentic layer on top of it: Antioch lets AI agents reason about robotic systems the same way coding agents reason about a codebase. Onboard a stack, calibrate high-fidelity digital twins, compose dynamic environments, run thousands of scenarios in parallel, and iterate, all without touching hardware. If you're an exceptional engineer who wants to work on this: antioch.com/careers If you're building in robotics/autonomy and want to speed up your development and testing, shoot me an email at alex@antioch.comVilli Iltchev reposted thisWe’ve raised $8.5M in seed funding, valuing Antioch at $60M, to build the simulation platform for physical autonomy. The raise was led by Category Ventures and A*, with participation from MaC Venture Capital, BoxGroup, Icehouse Ventures, Abstract, and angels. Antioch brings robotics and autonomy development entirely into simulation. Teams build, iterate, and test their full autonomous stacks in a single platform, accelerated by agents with native understanding of 3D environments. We're doing for physical autonomy what coding agents have done for software development. When the entire development loop lives in simulation, engineers and agents can reason about a robotic stack the same way they reason about a codebase: run it, observe the result, and iterate. The funding will go toward further expanding the engineering team, the core simulation infrastructure, and our new agentic framework. TechCrunch covered the raise and what we’re building: https://lnkd.in/g_UFyD8K If you’re an exceptional engineer interested in redefining autonomy development, we’d love to chat: antioch.com/careers.
-
Villi Iltchev shared thisThrilled to be supporting Antioch on their mission to build the simulation platform to bring testing and validation for robotics and autonomy from the physical world into software. Incredible team!Villi Iltchev shared thisWe’ve raised $8.5M in seed funding, valuing Antioch at $60M, to build the simulation platform for physical autonomy. The raise was led by Category Ventures and A*, with participation from MaC Venture Capital, BoxGroup, Icehouse Ventures, Abstract, and angels. Antioch brings robotics and autonomy development entirely into simulation. Teams build, iterate, and test their full autonomous stacks in a single platform, accelerated by agents with native understanding of 3D environments. We're doing for physical autonomy what coding agents have done for software development. When the entire development loop lives in simulation, engineers and agents can reason about a robotic stack the same way they reason about a codebase: run it, observe the result, and iterate. The funding will go toward further expanding the engineering team, the core simulation infrastructure, and our new agentic framework. TechCrunch covered the raise and what we’re building: https://lnkd.in/g_UFyD8K If you’re an exceptional engineer interested in redefining autonomy development, we’d love to chat: antioch.com/careers.
-
Villi Iltchev reposted thisVilli Iltchev reposted thisPodcasts are one of the richest sources of expert knowledge on the internet. A single Lex Fridman or Dwarkesh Patel episode can contain dozens of substantive claims about people, technologies, and organizations — but it's all locked inside hours of audio. You can't query any of it. You can't cross-reference what two different guests said about the same topic. In this post, we'll build a CocoIndex pipeline that turns YouTube podcast episodes into a queryable knowledge graph. The pipeline downloads audio, transcribes with speaker diarization, uses an LLM to extract structured statements and entities, resolves duplicates across episodes, and stores everything in SurrealDB as a graph. We use CocoIndex and Pydantic to build the pipeline. CocoIndex is a data indexing framework for building incremental data transformation pipelines — it tracks what's been processed, so re-running the pipeline only processes new or changed episodes. CocoIndex makes it exceptionally easy to build a knowledge graph without writing Cypher, and easy to add/remove any podcast incrementally. The full source code is available at CocoIndex Examples -- https://lnkd.in/g9QgHeh8 ⭐ Star the repo if it is helpful! #AI #LLM #CocoIndex #Podcast #MultiModal #EnterpriseAI #DataInfrastructure
-
Villi Iltchev reposted thisVilli Iltchev reposted thisMore than excited to announce Uncork Capital's investment in Soma Energy, alongside my good friends, Villi Iltchev at Category Ventures and Semil Shah at Haystack. There is increasing and well-founded concern that the world will not be able to produce enough electricity to support the expected insatiable demand of AI data centers over the next decade (we're expected to build 5x the current number of data centers in the next ten years). Athanasios Caramanolis, the Founder & CEO of Soma, and his team is building an important part of the solution. I share some thoughts about why we're so excited to partner with Ath. Uncork's view on seed stage investing is straightforward: we back exceptional teams, building cutting edge products, for big ass markets. Soma checks all three. A world-class team with direct, hard-earned experience, leveraging modern AI to tackle a market constraint that is both urgent and inevitable. https://lnkd.in/g2tyAR3PInvesting in Soma Energy: Making Every Megawatt Count - Uncork CapitalInvesting in Soma Energy: Making Every Megawatt Count - Uncork Capital
-
Villi Iltchev shared thisThrilled to be supporting the Soma Energy team on their mission to bring AI and automation to datacenters and power producers and unlock capacity and optimize power needs and cost. This is one of the biggest challenges and opportunities for the next decade.Villi Iltchev shared thisI spent my 20+ year career optimizing electricity to deal with grid constraints, interconnection delays, and delivering electrons more efficiently. I was struck by how challenging the problems were for large energy users while at AWS. We found ways through it but the underlying problem never went away. It got bigger. But from the inside, I had seen exactly what solving it would require. Today I'm proud to share that Soma Energy has raised $7 million in seed and pre-seed funding to tackle this problem. We built the platform I wish had existed when I was at AWS: one that connects generation, storage, and demand into a single control layer, so data centers can get to power in months instead of years and power producers can dispatch and trade with real precision. Thank you to our investors: Category Ventures, Uncork Capital, Haystack, RRE Ventures, TO VC, Panache Ventures, and Walter Kortschak. We are especially grateful to our customers and partners who believe in our mission, and to Mario Souto, Henrique Helfer Hoeltgebaum and the whole Soma Energy team. Also special thanks to Chris Neumann and Amir Rustamzadeh. Press release: https://lnkd.in/gttabY-pSoma Energy Emerges from Stealth with $7 Million in Funding to Unlock Power for the AI EconomySoma Energy Emerges from Stealth with $7 Million in Funding to Unlock Power for the AI Economy
-
Villi Iltchev liked thisExcited to see this effort launched 🚀! From 𝐜𝐨𝐧𝐭𝐢𝐧𝐮𝐨𝐮𝐬𝐥𝐲 𝐜𝐡𝐚𝐧𝐠𝐢𝐧𝐠 𝐮𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐝𝐚𝐭𝐚 to a 𝐊𝐚𝐟𝐤𝐚 stream — with CocoIndex and StreamNative. Many agent stacks today are built around periodic snapshots of their knowledge sources. For long-running agents that may execute for hours at a time, a snapshot captured at the start of the run quickly drifts out of sync with the underlying data. The combination of CocoIndex and Kafka takes a different approach: it treats the knowledge layer the same way operational data has been handled for years — as a stream of change events rather than a snapshot to be re-read. Drives, repos, design files, wikis, PDFs, and file shares — the unstructured data that has traditionally lived outside the streaming world — can be published to the same event backbone that already carries orders, clicks, and CDC traffic. The benefits show up in several places: • More efficient AI workloads. Embeddings, retrievals, and agent context are refreshed only when something has actually changed, which reduces redundant work and improves freshness at the same time. • A single change reaches every consumer. A commit, a renamed Drive document, or a Notion edit can update the vector index, notify an agent, update search, feed a Flink job, and land in a BI tile — without any of those systems needing to know about each other. • Easier extensibility. A new agent, a rebuilt RAG layer, or a compliance tool can be added as another subscriber to the topic, with the log providing replay so it sees historical changes the same way it sees new ones. • Better auditability. Each change consumed by an agent is durably recorded with offsets and timestamps, which makes it possible to answer questions like “did the agent see the updated policy before it acted?” with concrete evidence. • A stable contract over time. The change-event schema on the topic provides a stable interface between sources and consumers. Detectors, sources, and models can evolve independently while the wire format stays consistent. A static unstructured knowledge graph is rebuilt on a schedule, drifts between rebuilds, and pushes freshness logic into each consumer. A stream of unstructured change events stays current by construction and gives every consumer — agents, indexes, analytics, auditors — a shared view of what has changed. Most AI stacks today are built around the former; production agents tend to benefit more from the latter. CocoIndex is excited to build with infra friends at the ecosystem to provide a robust stack for the next generation of AI stack.Villi Iltchev liked this🚀 From 𝐜𝐨𝐧𝐭𝐢𝐧𝐮𝐨𝐮𝐬𝐥𝐲 𝐜𝐡𝐚𝐧𝐠𝐢𝐧𝐠 𝐮𝐧𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞𝐝 𝐝𝐚𝐭𝐚 to a 𝐊𝐚𝐟𝐤𝐚 stream — with CocoIndex and StreamNative Kafka's contract is famously simple: message in, message out. But the messages going in have always been the structured kind — orders, clicks, CDC off Postgres. Meeting notes, codebases, design files, wikis, PDFs? Those have lived in a parallel universe of vendor webhooks, nightly batch jobs, and one-off ETL nobody wants to own. We just closed that gap. With the new connector, you declare a Kafka topic as a target the same way you'd declare a Postgres table or a vector index. CocoIndex treats dynamically-changing unstructured assets as 𝐟𝐢𝐫𝐬𝐭-𝐜𝐥𝐚𝐬𝐬 𝐂𝐃𝐂 𝐬𝐨𝐮𝐫𝐜𝐞𝐬 and emits clean key/value change events — upsert, delete, no-op — incrementally, with no producer loop or "did I already publish this row?" bookkeeping. The post walks through a ~60-line example: a folder of CSV files watched live, each row published as a JSON message to a StreamNative-hosted topic. Edit a cell → exactly one message lands on the topic. Add a row → one new message. Delete a file → every row tombstoned. Why it matters for AI workloads: ★ Embeddings, retrievals, and agent context refresh only when something actually changes ★ One change event reaches every consumer — vector index, search, Flink, BI, agents ★ Replayable history, durable audit trail, stable schema contract over time Most agent stacks today re-index the wiki overnight and hope for the best. Production agents tend to do better on a stream of change events than on a snapshot read over and over. 👀 Read the post → https://lnkd.in/gHDeynwk 🌟Star the repo if it saves you an afternoon → https://lnkd.in/ezksH98y #Kafka #StreamingData #AI #DataEngineering #OpenSource #EnterpriseAI #StreamNative #AIEngineering #Agents #Realtime #UnstructuredData #DataInfrstructure
-
Villi Iltchev liked thisVilli Iltchev liked this7 years ago Job van der Voort and I started Remote. I was about to become a dad and between starting Remote and planning for a baby I realized shaving was sort of secondary.
-
Villi Iltchev liked thisSuper excited to finally share 𝐂𝐨𝐜𝐨𝐈𝐧𝐝𝐞𝐱 𝐯1 ! After 50 𝐫𝐞𝐥𝐞𝐚𝐬𝐞𝐬 𝐢𝐧 𝐯1 𝐚𝐥𝐩𝐡𝐚, together with 70 𝐜𝐨𝐧𝐭𝐫𝐢𝐛𝐮𝐭𝐨𝐫𝐬 since v0 launch. It is a fundamental redesign of how you write incremental data pipelines — built from a year of watching what people actually wanted to do with CocoIndex and building in the space. CocoIndex V1 is 𝐛𝐮𝐢𝐥𝐭 𝐟𝐨𝐫 𝐀𝐈 𝐞𝐧𝐠𝐢𝐧𝐞𝐞𝐫𝐬 𝐚𝐧𝐝 𝐚𝐠𝐞𝐧𝐭 𝐛𝐮𝐢𝐥𝐝𝐞𝐫𝐬 — people building coding intelligence, context, RAG, memory, knowledge-graph that live agents depend on. At GTC 2026, Jeff Dean and Bill Dally named a bottleneck that’s about to reshape every piece of infrastructure around AI. 𝐀𝐠𝐞𝐧𝐭𝐬 𝐫𝐮𝐧 𝐫𝐨𝐮𝐠𝐡𝐥𝐲 50𝐱 𝐟𝐚𝐬𝐭𝐞𝐫 𝐭𝐡𝐚𝐧 𝐡𝐮𝐦𝐚𝐧𝐬, 𝐛𝐮𝐭 𝐭𝐡𝐞 𝐭𝐨𝐨𝐥𝐬 𝐭𝐡𝐞𝐲 𝐫𝐞𝐥𝐲 𝐨𝐧 𝐰𝐞𝐫𝐞 𝐛𝐮𝐢𝐥𝐭 𝐟𝐨𝐫 𝐡𝐮𝐦𝐚𝐧 𝐬𝐩𝐞𝐞𝐝. Data infrastructure is one of those tools, and it matters beyond inference. An agent reasoning over a codebase, a conversation graph, a document corpus, or a stream of events needs that data fresh, organized, and cheap to query — not just on the first call, but throughout the run. 𝐓𝐡𝐚𝐭 𝐡𝐚𝐬 𝐚𝐥𝐰𝐚𝐲𝐬 𝐛𝐞𝐞𝐧 𝐂𝐨𝐜𝐨𝐈𝐧𝐝𝐞𝐱’𝐬 𝐯𝐢𝐬𝐢𝐨𝐧. 𝐕1 𝐦𝐚𝐤𝐞𝐬 𝐢𝐭 𝐭𝐡𝐞 𝐫𝐢𝐠𝐡𝐭 𝐬𝐡𝐚𝐩𝐞 𝐟𝐨𝐫 𝐚𝐠𝐞𝐧𝐭-𝐞𝐫𝐚 𝐰𝐨𝐫𝐤𝐥𝐨𝐚𝐝𝐬: the same incremental, state-driven guarantees, but now expressive enough to cover the pipeline shapes agents actually produce — entity resolution, clustering, multi-phase reduction, per-tenant topologies, conditional targets beyond embeddings and all. Every pattern in the examples gallery is something a long-horizon agent might want to run itself, and have its outputs become fresh source data for the next agent — without a human babysitting the job. Teams that take this on seriously typically allocate 10 – 20 engineers for at least six months to land the first production-worthy version — and then keep paying for maintenance indefinitely as sources, targets, and schemas evolve. CocoIndex ships all of this in the engine, so the code you write is the pipeline itself, not the scaffolding around it. Take a look at the announcement here, and keep us on your feedback! blog: https://lnkd.in/gdcvs7Nh star us on github: https://lnkd.in/gb72VtGh Thank you so much for your continuous support on us, we are so excited to be on this together!!Villi Iltchev liked this🚀 CocoIndex V1 is live. It's the incremental context engine we've been rebuilding for the agent era. Long-horizon agents run roughly 50x faster than humans. They write code, record their decisions, and update their own context as they go. Meanwhile, most of the data infrastructure they rely on is still built around a human's morning dashboard. That cadence doesn't hold up when agents are in the loop. By the time the nightly rebuild finishes, the context is already behind. CocoIndex has always had one job: keep derived data in sync with source data, without redoing work that's already been done. V1 is a fundamental redesign, shaped by a year of watching what teams actually build with Coco. Here's what that looks like in practice: 🔹 A live index of your codebase, a knowledge graph from your meeting notes, or a vector store over your docs can stay fresh without you having to think about it. Edit a paragraph and only the affected chunks get re-embedded. Delete a file and its rows go away on their own. Swap the embedding model and the column re-syncs. 🔹 You can stand up a continuously fresh context pipeline in an afternoon instead of planning a quarter of work around it. Point CocoIndex at your source, describe what you want in the target, and it keeps Postgres, VectorDB, Neo4j, Kafka, S3, or even a folder on disk in that shape as things change. 🔹 The pipelines agents actually produce are finally expressible end-to-end: entity resolution across sessions, clustering, multi-phase reduction, per-tenant topologies, conditional targets. If you can write it in Python, you can run it incrementally. 🔹 Multi-tenant setups stop being a separate project. One Postgres schema per tenant, one Kafka topic per category, one knowledge DB table per entity type, all driven by a live registry. When a tenant or category goes away, CocoIndex cleans up after it. 🔹 Going from `pip install` to a running pipeline takes minutes. The engine is embedded, so there's nothing to stand up alongside it. You prototype locally and ship the same code to production. The mental model is still the clearest way to describe what CocoIndex does. You declare what the *target* should look like as a function of the *source*, and the engine works out the transitions. Think React, but for data. Building a production-worthy incremental engine in-house usually takes 10 to 20 engineers and at least six months, and then keeps asking for maintenance forever as sources, targets, and schemas evolve. CocoIndex absorbs that work into the engine, so what you write is the pipeline itself, not the plumbing around it. Huge thanks to the 70+ contributors who have shaped CocoIndex since the v0 days, and to everyone who lived on the 50 alpha releases with us on the road to V1. - Read the announcement: https://lnkd.in/gi3M3hrV - Star on GitHub: https://lnkd.in/ezksH98y #AI #AIAgents #DataEngineering #OpenSource #RAG #Python #EnterpriseAI #AIInfrastructure
-
Villi Iltchev liked thisVilli Iltchev liked thisProud dad moment! Last week, my wife, son, and I spent spring break in New York City taking in the sights and sounds. But the true highlight was spending time with my daughter, who set a goal a year ago to graduate from #UVA and build a life in #NYC. Growing up, I was given limited choices and never wanted that to be her story. She has taken full advantage of every opportunity in front of her and has quickly shown a level of independence, resilience, grit and maturity that I deeply admire. She achieved her goal (many more to come) and reminded me that no dream is out of reach when you pair vision with hard work. Asha, I am incredibly proud of you and cannot wait to see what you conquer next. Anita Gupta #prouddad
-
Villi Iltchev liked thisVilli Iltchev liked thisCocoIndex coming to East Bay Rust 🦀 Meetup! All about Rust with friends at Zed Industries, Convex and all. CocoIndex CTO George He sharing the fun stuff about building Rust data engine. The usual pattern for this kind of library — NumPy, PyTorch, Polars — puts the heavy compute on the native side and lets Python orchestrate. We built CocoIndex the other way around. User compute runs in Python (embeddings, LLM calls, parsing). The Rust core handles the less glamorous parts: change detection, scheduling, persistence, state. Most of the interesting engineering lives at the seam between the two. A few of the things we covered: • Bridging async across two event loops (Tokio and asyncio), and the one rule that makes both directions tractable • GIL discipline beyond the obvious — what the compiler catches, and what it doesn't • A fatal shutdown-time abort and what actually caused it • One generic Rust primitive that ended up powering two completely unrelated problems — batching db writes and GPU/embedding calls from user code Thanks to Esther Trapadoux Zed Industries for having us! More on CocoIndex: https://lnkd.in/ezksH98y #Rust #Python #OpenSource
Experience & Education
-
Category Ventures
View Villi’s full experience
See their title, tenure and more.
or
Languages
-
Bulgarian
-
-
Greek
-
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content