About
I spent 7 years in the OpenShift and DevTools trenches, co-founding Shipwright (now a…
Activity
2K followers
-
Shoubhik Bose shared thisRunning models is one thing. Deploying them continuously without violating your security team's policies is another.Shoubhik Bose shared thisDefense AI has a problem most vendors won't say out loud. You can verify who's authorized to run a model, but not whether the model running today matches the one your security team approved last week. Supply chain tampering lives in that gap. Runtime drift opens another. A few weeks ago Jozu was deemed Awardable by the DoD Platform One Solution Marketplace, recognized for our work on AI assurance for mission-critical environments. What we cover: cryptographic attestation of every model, agent, and MCP server before it runs; policy enforcement during execution; and a tamper-evident audit trail for after. It runs in connected, on-prem, and air-gapped environments. This is our second federal awardable, after CDAO Tradewinds in February. Grateful to the P1 evaluators for the rigorous review, and to the team for shipping something serious enough to clear it. Government customers, our 5-minute pitch is live on the marketplace: https://lnkd.in/eJeTFNBc Press release: https://lnkd.in/enX6rycp
-
Shoubhik Bose reposted thisShoubhik Bose reposted this🚀 Today we're excited to announce Jozu Agent Guard! Over the last year our team has been heads down working on securing the ML supply chain, we've pioneered the ModelKit artifact standard to streamline development, built Jozu Hub as a central AI/ML registry, added comprehensive security scanning, cryptographic signing, tamper-proof audit logging, and rapid inference creation. Now, Jozu is expanding into the runtime environment with Jozu Agent Guard. Agent Guard is built to bring confidence to the rapidly evolving agentic world that we're transitioning into–allowing you to empower your agents to act, and trust that they don't act outside of the boundaries you set. Agents are powerful, and with that power comes risk. Unlike a human, they don't always see the full picture and simple, non-malicious actions can quickly turn into nightmare situations. Agent Guard was built to eliminate that risk. Learn more by clicking the link in the first comment ;)
-
Shoubhik Bose shared thisI'm speaking at the Airflow Summit 2025 in Seattle next week. Looking forward to meeting the lovely folks at Astronomer and the broader community!
-
Shoubhik Bose posted thisAfter spending the last 8 years dealing with developer tools, containers and Kubernetes, I'm moving on to start a new position as Technical Data Architect on the Data Development, Insights & Strategy team at Red Hat to help modernize how Red Hat builds data products to better inform operational decisions. 🎓
-
Shoubhik Bose shared thisThis is big! The Argo Project has graduated!Shoubhik Bose shared thisAs of a few hours ago, the #ArgoProject has been approved for graduation, after a record fast Cloud Native Computing Foundation (CNCF) TOC vote! It's been a long process with a few twists and turns, but the hard work from everyone in the project and the support from all the users have carried us through. Im incredibly grateful to be part of this awesome group of people and extremely thankful to my employer Intuit, who sees the value of #opensource and understands the value of investing in and supporting our communities. Im also very excited for the future of the project, the past years have only been the beginning and there's so much more yet to create and achieve together! A massive thanks to the TOC sponsors Lei Zhang, Cathy Zhang and Davanum Srinivas for all the support, guidance and patience through this! Of course, this wouldnt have been possible without the amazing group of maintainers: Rohit Agrawal, Zach Aller Leonardo Luz Almeida Saravanan Balasubramanian Chetan Banavikalmutt Simon Behar @xianlu bird Shoubhik Bose Remington Breeze Yi Cai @keith chong Alex Collins Michael Crenshaw Jann Fischer Dan Garfield Ravi Hari @hui kang Saumeya Katyal Kostis Kapelonis Pasha Kostohrys Edward Lee Alexander Matyushentsev @mikhail mazurskiy Vaibhav Page Andrii Perenesenko Hari kumar Rongali Regina Scott Ishita Sequeira Daniel Soifer Jesse Suen Yuan (Terry) Tang @william tam Daisuke Taniwaki Derek Wang Hong Wang @jonathan west @may zhang Tianchu Zhao And the maintaining companies: Akuity BlackRock Codefresh Intuit Red Hat Onwards and upwards! #cloudnative #argocd #kubernetes #devops #mlops #cicd #gitops
-
Shoubhik Bose shared thisThat was an exciting Office Hours on "The Path to GitOps" with Harriet Lawrence and Jaafar Chraibi . #KubeCon2022 #KubeConEU #GitOps #cloudnative
-
Shoubhik Bose shared this🥳Shoubhik Bose shared thisOpenShift GitOps out here winning awards! Read about how #RedHat #OpenShift #GitOps won the DevOps Dozen Award! https://lnkd.in/gPAN3rW8
-
Shoubhik Bose shared thisI'm hosting a virtual meetup on 1-min impromptu speaking! Feel free to hop in and give impromptu speaking a shot 👋 Checkout this Meetup with The Mouthfuls: https://lnkd.in/eY4YxixY #publicspeaking #leadership #communicate
-
Shoubhik Bose liked thisShoubhik Bose liked thisSolomon Hykes nailed something in a post this week: Skills are powerful, but they lack a clear packaging format. No signing. No versioning. No standard for how they get distributed, verified, or deployed. The replies are full of people asking the obvious next question, "where's the signed manifest? where's the provenance? how do I know this skill is safe to run?" This is exactly the problem we've been working on at Jozu, and it's why we shipped [kit unpack --as-skill] in the most recent KitOps release, v1.13.0. Skills are just another kind of AI artifact, they're bundles of instructions, scripts, and context that an agent loads at runtime. And we already know how to package AI artifacts properly: as OCI artifacts, with a manifest, a digest, signatures, and an SBOM. That's what a ModelKit is. KitOps has been doing this for models, datasets, code, and config for two years. Adding skills to the same packaging format wasn't a stretch — it was the obvious move. What that gets you: • 𝗢𝗻𝗲 𝗮𝗿𝘁𝗶𝗳𝗮𝗰𝘁, 𝗼𝗻𝗲 𝗱𝗶𝗴𝗲𝘀𝘁. A skill (or a bundle of skills, with the model and config it depends on) lives at a versioned, content-addressed reference in any OCI registry you already trust. • 𝗦𝗶𝗴𝗻𝗲𝗱 𝗮𝗻𝗱 𝘃𝗲𝗿𝗶𝗳𝗶𝗮𝗯𝗹𝗲. Cosign works out of the box. You know who built it and that it hasn't been tampered with. • 𝗸𝗶𝘁 𝘂𝗻𝗽𝗮𝗰𝗸 <𝗿𝗲𝗳> --𝗮𝘀-𝘀𝗸𝗶𝗹𝗹=𝗰𝗹𝗮𝘂𝗱𝗲-𝗰𝗼𝗱𝗲 drops it into the right place on disk for the agent harness you actually use. No app store. No copy-paste from a gist. No "curl | bash" energy. • 𝗦𝗮𝗺𝗲 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄 𝗳𝗼𝗿 𝘁𝗵𝗲 𝗿𝗲𝘀𝘁 𝗼𝗳 𝘁𝗵𝗲 𝘀𝘁𝗮𝗰𝗸. If your skill ships alongside a fine-tuned model or a RAG corpus, they all travel together as one artifact. The "app store of skills" path that Solomon flagged is going to generate a lot of noise. The boring answer, just reuse the packaging, signing, and distribution standards the cloud-native world spent a decade getting right. KitOps is open source and a Cloud Native Computing Foundation (CNCF) project. You can check it out at KitOps.org
-
Shoubhik Bose liked thisShoubhik Bose liked thisMost "AI agent security" tools solve a sliver of the problem. Sandboxes constrain where code runs. They don't govern what the agent does. AI gateways see network traffic. They're blind to stdio MCP servers, local file ops, and shell commands. Cloud guardrails filter model inputs and outputs. They don't govern what the agent does with the response. Three different blind spots — and an agent only needs one to do damage. Jozu Agent Guard closes all three: → Verifies artifacts before they execute (Cosign + OCI attestations) → Enforces policy on every agent action at runtime, including local ones → Tamper-evident, cryptographically chained audit logs → Runs on-prem, at the edge, and air-gapped — no cloud dependency Zero-trust for AI across the full lifecycle. Not just the parts that cross a network boundary. See the full comparison: https://lnkd.in/e4F_uYai
-
Shoubhik Bose liked thisShoubhik Bose liked thisKitOps v1.12 is our first step to secure packaging for agentic AI. KitOps now supports agent skills: run `kit init` in a directory with a SKILL.md and Kit will detect it, parse the frontmatter, and generate a Kitfile that includes it as a prompt layer you can unpack independently. This is groundwork for what's coming next: unpacking ModelKits as agent skills directly into tools like Claude Code and Cursor. The other theme in this release is flexibility for how real-world projects actually look. Most teams aren't working with everything in one place. They have open models from HuggingFace, training data in S3, and local agents they're iterating on. KitOps now handles all three: reference datasets in S3 buckets without duplicating them into packages, import HuggingFace datasets directly, and run `kit init` against remote HuggingFace repos to generate a Kitfile you can edit before importing. This gives you the same integrity guarantees regardless of where your artifacts live. We're grateful for the great contributions from four new community members. Open source works when people show up, and they did. Link to the full release notes and installation instructions in the first comment. #KitOps #CNCF #MLOps #OpenSource
-
Shoubhik Bose liked thisShoubhik Bose liked this𝐉𝐨𝐳𝐮 𝐋𝐚𝐮𝐧𝐜𝐡𝐞𝐬 𝐀𝐠𝐞𝐧𝐭 𝐆𝐮𝐚𝐫𝐝 𝐟𝐨𝐫 𝐀𝐈 𝐒𝐞𝐜𝐮𝐫𝐢𝐭𝐲 Jozu has introduced Agent Guard, a zero-trust AI runtime designed to secure AI agents, models, and MCP servers with built-in governance that agents cannot bypass. The solution ensures continuous policy enforcement, controls agent actions, and maintains tamper-proof audit logs addressing critical risks as enterprises scale agentic AI adoption. “The AI exhibited a pattern indistinguishable from a malicious insider: disable the monitoring, erase the logs, carry on like nothing happened.” — Brad Micklea, Co-Founder & CEO, Jozu Read more: https://lnkd.in/gsSrZCsz Want to position your brand in front of cybersecurity leaders driving AI adoption? Access our Media Kit- https://lnkd.in/d-vZH_3J #CyberSecurity #AI #AgenticAI #ZeroTrust #AISecurity #MCP #DevSecOps #Infosec #EnterpriseAI #B2BTech
-
Shoubhik Bose liked thisShoubhik Bose liked thisAfter four exciting, fulfilling years, I've decided to leave Chainguard. Since then, I've been taking some time off. This is the first time in my entire career I've had more than a weekend of time off between jobs. I just got back from a two week trip to Morocco with the family (one huge benefit of being married to a travel advisor! Lauren Cherney for all your travel needs!). And I've also spent a bunch of time building fun stuff in my now-copious spare time. I'm looking forward to figuring out what's next -- if you're working on something cool and think I'd be useful, let me know (no recruiters please!). But in the meantime I'm going to keep building cool stuff. I'll share some here as I build them, starting tomorrow until I run out of stuff to share :)
-
Shoubhik Bose liked thisShoubhik Bose liked thisAnthropic launched Managed Agents yesterday. I think of it as Heroku for Agents. That's great but it doesn't solve everything your org needs. Managed Agents solves the infrastructure overhead that keeps most teams from shipping agents to production. You define your agent and let Anthropic handle sandboxing, orchestration, and error recovery. But there is still a lot that your organization is responsible for... 1/ If you need to run agents on-prem, at the edge, or disconnected, Managed Agents won't help you. 2/ Managed Agents applies only to Claude agents running on Anthropic's cloud. Nearly all the organizations we talk to need multiple agents from multiple vendors. You're still responsible for providing teams consistent security and governance across all of those - including Claude. If you're in a regulated industry it gets more complex because you need audit-ready evidence, not just a session log. 3/ There's no supply chain verification, so no way to prove the model or MCP server your agent loaded is the one your security team approved. No cryptographic attestations. No customer-owned policy engine. The guardrails are coarse-grained: the GitHub MCP server exposes 91 tools. Most agent tasks only require a handful and giving it everything invites mistakes and security issues. You need a way to stop access to specific MCP tools, not just the whole MCP server. Managed Agents makes it easy to deploy agents which is great, but it doesn't answer the harder question: how do you secure and govern all your agents everywhere you need them to run? That's what we've built over the last three years at Jozu. #AIAgents #AgentSecurity #AI
-
Shoubhik Bose liked thisShoubhik Bose liked thisYour AI coding agent often has your machine access: SSH keys, cloud credentials, push-to-main, network access. The only thing between the agent and your infrastructure is a prompt. That's not a security model. We built Jozu Agent Guard to replace trust with enforcement. It's a secure runtime for AI agents: declare the agent, and Agent Guard runs it in a governed environment where actions are policy-evaluated. Declare the agent, not just the policies. Agent Guard introduces agent definitions: a declarative YAML format for the LLM, MCP servers, policies, and skills. Definitions compose through base references. Security teams publish a base agent with corporate policies baked in. Developers extend it for their projects. Policies are additive only: a derived definition cannot remove a policy module from its base chain. If security declares it, downstream developers can't strip it out. Every module type — models, MCP servers, policies, skills — can be pulled from OCI registries as signed, versioned ModelKit artifacts. The runtime is the enforcement boundary. Agent Guard boots a disposable microVM in under a second on developer machines and runs Kata containers on Kubernetes. By default, the VM gets nothing: no home directory, SSH keys, cloud tokens, or API keys. The agent starts with zero credentials. You opt in to what crosses the boundary. Only named credentials enter the VM, in-memory and never written to disk. If the agent needs GitHub and AWS credentials, that's all it gets. Kubeconfig, Azure tokens, SSH keys, and everything else stay outside. This is the difference between an agent with your access and one with its own. If a compromised agent or MCP server exfiltrates credentials, the blast radius is limited to what you granted. When the session ends, workspace changes sync back. The VM is discarded. Inside the runtime, an AI gateway gives the policy engine visibility into LLM traffic and MCP calls. Three policy kinds, all enforced locally. Policies are CEL expressions in YAML, distributed as signed OCI artifacts. ToolPolicy: evaluated on every tool call. Controls tool access and arguments. GuardrailPolicy: evaluated on inference traffic for content safety, PII detection, and prompt injection defense. ArtifactPolicy: evaluated at admission. Verifies signatures and provenance before artifacts load. Fails closed on missing data or evaluation errors. Global policies layer with workspace-scoped policies. They auto-sync from registries on every run, with optional cosign verification. Every decision is logged and synced to Jozu Hub when connected, or stored locally when not. Agent Guard supports Claude Code, Gemini CLI, OpenClaw, Codex CLI, and others. If your agents run with your full credentials and no declared boundaries, you're trusting the model with everything. Agent Guard gives each agent exactly what it needs and nothing more. https://lnkd.in/eDzbHFSC
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
Licenses & Certifications
Volunteer Experience
-
Mentor
Massachusetts Institute of Technology
- Present 6 years 9 months
Science and Technology
I provide industry-grade code review of student's assignments for the "Performance Engineering of Software Systems" class
-
Publications
-
Service Crawling in Cloud Computing
Springer
The cloud service user of a single service relies on specific service capabilities to query relevant data. It can be difficult to consistently query since data is distributed across multiple services. In this paper, we are going to develop an algorithm to crawl the services within a cloud service inventory invoking fetch capabilities and following Internet protocol addresses among the resources assembling retrieved data in a centralized query enabled indexing service. Crawling techniques are…
The cloud service user of a single service relies on specific service capabilities to query relevant data. It can be difficult to consistently query since data is distributed across multiple services. In this paper, we are going to develop an algorithm to crawl the services within a cloud service inventory invoking fetch capabilities and following Internet protocol addresses among the resources assembling retrieved data in a centralized query enabled indexing service. Crawling techniques are being used for pre-cache information [1] that a user needs for post processing information at current available resource(s) for improving user latency to safe requests.
Other authors -
Honors & Awards
-
Winner - Random Hacks of Kindness ( Bangalore ) 2012
RHoK
-
Finalist - IBM The Great Mind Challenge 2011
IBM India
-
Placement coordinator
Netaji Subhash Engineering college
I worked with the Placement department of Netaji Subhash Engineering College to facilitate the Campus recruitment process.
Key duties performed:
- Coordinate campus drives of IBM , MQuotient, TCS, Accenture.
- Data collection and management of students's academic data.
- Develop software to enable updation of students' data.
- Conduct regular KT ( Knowledge Transfer workshops) on commonly used technologies like RSS, Yahoo Pipes, Git , etc )
-
Runners up : Inter-college Technical Quiz at Avenir 2011
Phoenix - Official Tech club of Netaji Subhash Engineering College , Kolkata
Organizations
-
Boston One Dragon Boat team
Paddler
- Present
Recommendations received
5 people have recommended Shoubhik
Join now to viewOther similar profiles
Explore more posts
-
Anderson Luiz Mendes Matos
Today at #RedisReleased Toronto I explored how Redis is evolving far beyond what most of us still picture it as. From massive performance gains and multi-cloud active-active deployments to data aggregation, extreme resiliency, and real-time load management without bottlenecks, Redis is a complete data platform. It was a privilege to see, first-hand, feedback from TD and OrderGrid and how Redis is empowering their next steps, including AI transformation and adoption. Add reliable messaging, deep instance insights and even the ability to chat with your data, and the message is very clear: this isn't just about caching for a long time, now. Redis is empowering global enterprises to run faster, smarter, and more resilient than ever. And you? Are you still using Redis as a cache? What's your favorite Redis feature (or the one that surprised you the most)? #RedisReleased #Redis #DataEngineering #Performance #Cloud #HighAvailability #TorontoTech #AI
-
Kamran Ali
July 19, 2019, an outside individual gained unauthorized access about Capital One credit card customers This event affected ~100 million individuals in the United States and ~6 million in Canada How to beef up Authentication Process Solution: Two-Factor Authentication (2FA) It's a Two-Step Verification Process - First, you provide something you know (your password) - Second, you provide something you _have_ (a time-based code from your phone or a physical key). Here’s how it works, 1. User enters username and password 2. System validates credentials against database 3. If valid, system generates a unique verification code 4. Code is sent to user's registered device (SMS/email/app) 5. User enters the code to complete authentication 6. System grants access only after both factors are verified 🤔 over to you, what to do if 2FA device is lost? Let me know in the comments! 👇 --- You've got System Design skills but showcasing them in 60 minutes interview is a completely different challenge for e.g. where to start? How much time to spend on HLD vs LLD etc Let me share a FREE 7-day email course that provides a step-by-step framework for acing your 60-minute System Design interview 📣 Join 1700+ subscribers and sign up for the course to learn time allocation to demonstrate full range of design skills (link in comments 👇)
7 Comments -
James Cunningham
Risks of SaaS Data Storage Outside Canada ---------------------------------------------------------- Canadian companies and organizations must be aware of where their data is stored. Certain industries—such as medical, dental, and healthcare—are legally required to ensure that all files and information are stored exclusively within Canada. Many SaaS applications are hosted on cloud infrastructure located outside of Canada. Even when a vendor claims their primary servers are Canadian-based, backup, disaster recovery, or support systems are often hosted in other countries. SaaS providers are not obligated to proactively disclose where customer data is stored. In many cases, vendors provide vague, incomplete, or incorrect information, and frontline representatives often do not know the full details of their data residency practices. The safest approach is to require the SaaS provider to formally certify in writing—signed by a senior executive—that all data, including backups, remains within Canada. Ultimately, the responsibility and liability rest with your organization, not the SaaS vendor. The fact that a SaaS product is sold or marketed in Canada does not mean it complies with Canadian data residency laws. Failure to verify compliance can expose your organization to legal and regulatory risk.
-
Darryl R.
This past Saturday, I presented a deep dive on AWS Lambda Durable Functions at the AWS London Ontario User Group meetup. The presentation explores how Durable Functions work and how they compare to Step Functions and Temporal for workflow management. I also built a demo application, available on GitHub, so you can try it yourself. Links to our User Group YouTube channel and Meetup page are in the comments. Thank you very much to Ngwa Bandolo Bobga Cyril for organizing this session and all the previous ones! #AWSCommunity #AWSUserGroupLondonON https://lnkd.in/ePkmPQKG
5 Comments -
Tomasz Spłuszka
🚀OpenText just launched the powerful OpenText AI Data Platform - a unified, open data platform built for enterprise AI: 🔗 Zero-copy access – keep data where it is, no duplication 🔒 Enterprise-grade governance & security built in 🤖 Powers specialized AI agents (Aviators) for your industry ☁️ Multi-cloud, on-prem, BYOM, deep integrations (ERP, CRM, Databricks, etc.) 🧑💻 Aviator Studio (no-code agent builder) Real, trusted, governed AI that turns your private data into business impact – securely and at scale. https://lnkd.in/ejyBtcYR #EnterpriseAI #GenAI #OpenTextAviator #DataPlatform #DigitalTransformation
-
Nikhil Sharma
𝗛𝗼𝘄 𝗣𝗼𝗹𝗶𝘁𝗶𝗰𝘀, 𝗡𝗼𝘁 𝗣𝗵𝘆𝘀𝗶𝗰𝘀, 𝗣𝗼𝘄𝗲𝗿𝘀 𝗢𝘂𝗿 𝗧𝗶𝗺𝗲 𝗭𝗼𝗻𝗲𝘀 I have always been curious about how programming languages and time libraries manage time zones. British Columbia's recent decision to observe Daylight Saving Time (DST) year-round finally prompted me to investigate further. Ideally, time zones would be governed by physics - the Earth's rotation - but in reality, they are dictated by politics. While political maneuvering often leads to chaos, our clocks, servers, and phones continue to operate seamlessly throughout the year. This reliability stems from a shared foundation: the 𝗜𝗔𝗡𝗔 𝗧𝗶𝗺𝗲 𝗭𝗼𝗻𝗲 𝗗𝗮𝘁𝗮𝗯𝗮𝘀𝗲. This volunteer-maintained database powers the timezone logic for PHP, Node.js, Linux, PostgreSQL, and most of the internet. When an application specifies a zone like "America/Toronto" the runtime consults this database to resolve the current offset. It contains hundreds of government rules, DST transitions, and historical data that adapt to political changes automatically. 𝗙𝘂𝗿𝘁𝗵𝗲𝗿 𝗥𝗲𝗮𝗱𝗶𝗻𝗴: 1. 𝘏𝘰𝘸 𝘢𝘳𝘦 𝘵𝘪𝘮𝘦 𝘻𝘰𝘯𝘦𝘴 𝘤𝘰𝘰𝘳𝘥𝘪𝘯𝘢𝘵𝘦𝘥? – ICANN Blog: https://lnkd.in/eyYGq_GJ 2. 𝘉𝘳𝘪𝘵𝘪𝘴𝘩 𝘊𝘰𝘭𝘶𝘮𝘣𝘪𝘢 𝘵𝘪𝘮𝘦𝘻𝘰𝘯𝘦 𝘶𝘱𝘥𝘢𝘵𝘦 – GitHub Commit: https://lnkd.in/et6PmFnz #tech #insight
-
LearnKube
Master Kubernetes with LearnKube's Advanced Kubernetes workshop! What should you expect? - Learn how to architect and design clusters from the ground up (in the cloud or on-prem). - Explore the Kubernetes internal component and how the system is designed with resiliency in mind. - Deep-dive into the networking components and observe the packets flowing into the cluster. - Hands-on labs to test the theory with real-world scenarios! - And more. The next course starts in September: https://lnkd.in/ge4aYVZq We also run in-person courses and private training: https://lnkd.in/g9N3m-Xr
-
Manikyam Thukkapuram
Excited to see the momentum as Quadrant Technologies expands its presence in Canada. As part of this journey, we are engaging with industry leaders in Vancouver to explore how organizations can build a stronger data foundation on Microsoft Azure to unlock AI innovation. Personally excited to be part of Quadrant’s Canada expansion, connecting with customers and partners through workshops that showcase real-world transformation experiences and modernization strategies. #Azure #AIInnovation #DataPlatform #CloudModernization #Canada #QuadrantTechnologies Vamshi ReddyKaivalya HanswadkarPrasanth Venkata Satya TammirajuPrakash KarakaAnanda Kumar DeyDerek WolcoskiPhilip PankovJay PatelBrian Beyerbach
1 Comment -
Ishaan Bhardwaj
Observability adoption benefits from a "middle-out" approach, as noted by Michael Woodside, Director of Global DevOps at Pacvue. Equipping engineers, developers, and product managers with observability tools can lead to organizational influence in both directions. The outcome is a culture where decisions are informed by data-driven insights rather than intuition alone, fostering improved decision-making processes across the organization. Watch Andre Scott's conversation with Michael on demand at https://bit.ly/3M70tB6
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More