Praveen

Developer, Tech and Finance Enthusiast | Forward Deployed Engineer @PromptQL

[Praveen @praveenweb]

A year of deploying this inside Fortune 500 teams taught me one thing. The AI gets good the moment the whole team can teach it in one place. Proud of the launch today🚀

[Tanmai Gopal @tanmaigo]

We raised $136M to kill Slack.

Introducing PromptQL: The first AI version of Slack.

Here’s how it works:

[Rajoshi @rajoshighosh]

Meet PromptQL: Multiplayer AI workspace for your team --> outputs a shared wiki. Shared context builds as you work; exactly how @karpathy describes it. Teach a concept once--> the system learns --> everyone benefits. We've been using it internally and it's been a game-changer for keeping up with AI-speed work. https://t.co/KuqiONKHj5

[PromptQL @PromptQL]

Going to @DevNetwork_'s #AIDevSummit? Don't miss @praveenweb's session "Path towards 100% Accurate & Repeatable Data Agents for AI"

See how #PromptQL's agentic query planning eliminates the goal drift and multi-step errors that break traditional approaches.

[Praveen @praveenweb]

@JiquanNgiam Hey @JiquanNgiam,
Read your blog (trending on HN 😄)

Decoupling query planning from execution is an architectural decision that we chose to get to 100% accuracy with AI. I did some write up around how we do this at PromptQL.

Would love your thoughts!
https://t.co/koRb9Ofzda

[Praveen @praveenweb]

@sama @AravSrinivas How about an API for SearchGPT? Perplexity's API seems to use an inferior model and there is a big opportunity.

[Praveen @praveenweb]

@TejasKumar_ Search doesn’t work for Reminders app on my Mac. It is just so frustrating to see basic functionality broken.

[Sam Hogan 🇺🇸 @samhogan]

LLM chat interfaces need to support "branching" so it's possible to create a new thread with the *current* chat context to ask tertiary questions without polluting the context of future questions

[Tanmai Gopal @tanmaigo]

A simple measure of real-world usability for an AI system (apart from accuracy ofc) is repeatability

Repeatability ⬆️ => Trust ⬆️ => Adoption ⬆️

As a test, used Claude to get top 5 priority tickets from last 5, 10 and 20 recent open tickets

It's bad and gets worse.

A 🧵:

[Praveen @praveenweb]

Human-in-the-loop systems are essential for Agentic AI applications where sensitive decisions require oversight. Today, I built a Customer Support Assistant using @HasuraHQ PromptQL showing a controllable agentic flow.

This AI Assistant has access to customer invoices📊 and support ticket 📨 data.

Here’s how it works:
💡 When a refund action is triggered, PromptQL doesn’t just execute—it creates a query plan to gather context.
🙋‍♂️ It then pauses for manual approval or denial before proceeding.

This manual intervention makes it safer for data manipulation in agentic workflows.

How are you using human-in-the-loop systems in your AI applications? What are the different use cases? Would love to know what everyone else is building.

#HumanInTheLoop #AgenticAI #AgenticRAG

[Praveen @praveenweb]

As an Open Source maintainer, managing GitHub projects can be overwhelming 🐱

I built an AI Assistant with PromptQL by @HasuraHQ to analyze GitHub repo data. I chose the popular @nextjs repo from @vercel which had about ~3k open issues and used PromptQL to help prioritize issues.

Here’s the magic:
When I asked it to help prioritize the 10 most recent issues, PromptQL retrieved the data and created a query plan to:

1️⃣ Analyze GitHub issues and comments and get the title/body content.
2️⃣ Use a language model to classify the issue priority as high, medium and low.
3️⃣ Store the results in an artifact and summarize the findings.

The assistant gave me actionable insights faster than I ever could on my own!

If you’re an OSS maintainer like me, juggling hundreds of issues, this is definitely going to help with productivity. I work with multiple repositories across Hasura and would love to experiment with that as a next step.

💡 Check out the demo and let me know: What would YOU ask PromptQL to do for your GitHub projects?

#PromptQL #AgenticRAG #AgenticAI #GitHub

[Praveen @praveenweb]

Exploring ChatGPT Search these past few days. Here's my early understanding.

- Powered mainly by Bing [Source: OpenAI Reddit AMA]
- Doesn’t seem to pull content from forums (Reddit/Quora)
- Bing prioritizes multimedia content (images/videos)

Takeaway: Ranking high on Bing might be key if you want your site or product to show up in ChatGPT citations!

[PromptQL @PromptQL]

Hot take: AI Assistants are failing us

Despite the buzz, closed-domain AI assistants are falling short. Without reliable, context-aware responses, they’re not ready for serious business use.

Where AI Assistants Fail

Here’s a scenario from a well known sales assistant that’s out there today:

📊 User Query: “What’s the length of my average sales cycle?”
➡️ Assistant Response: “I calculated the average sales cycle length for your opportunities, but there are no results to show.”

The assistant can’t perform a computation. Why? Let’s break it down.

🛑 The Issue: Closed-domain AI assistants rely heavily on search-first RAG methods, making them unsuitable for high-trust applications.

Consider a task like “Find all emails from last week that need follow-ups.” A search-based AI might skip important messages if they lack specific keywords, leaving critical follow-ups unnoticed. When this incomplete data is passed to the language model, the result is unreliable, making these assistants ill-suited for nuanced business queries.

✅ The Solution: Agentic query planning. Instead of rigid keyword search, assistants should gather all relevant emails and then use an LLM to classify follow-ups—just as a person would—ensuring accuracy.

That’s why our AI lab built PromptQL - an agentic, data access API for your AI!

Here’s a look at what we’ve been up to ⤵️

[SingleStore @SingleStoreDB]

Another day, another awesome webinar you don't want to miss!

Please join us on October 24th at 10am PDT for a detailed webinar on how to build secure, scalable data access APIs using #Hasura and #SingleStore.

💡 Learn best practices for securing your APIs, optimizing for performance, and handling complex queries at scale.

👉 RSVP today: https://t.co/3szyZsd9Zh

#DataSecurity #RealTimeData #SecureAPI

[Praveen @praveenweb]

@VicVijayakumar @TheTrentHarvey Tinkering with cPanel was a good deterrent. Not sure how it is setup these days. But definitely not the easiest way.

[Praveen @praveenweb]

Google's own apps (Slides/Docs) do not support uploading webp images, an image format that they default to in their own browser while saving images 🤦‍♂️

[Jeff Cotrupe @jeffcotrupe]

Building Secure, Scalable Data Access APIs with @HasuraHQ and @SingleStoreDB
Join Anuja from #SingleStore and Praveen from #Hasura to learn how to create robust and scalable APIs.
October 24, 2024 | 10:00-11:00am PDT
JOIN US! Register now:
https://t.co/xtSCSeyNsm

[Sam Julien @samjulien]

“DevRel is dead”

Okay then why is literally every AI DevTool looking for dev advocates right now lol

[PromptQL @PromptQL]

We're hosting a live webinar on how to use the Hasura Elasticsearch GraphQL connector (13 Aug)!

The Agenda:

🧙 Generate instant APIs
🔮 Advanced query capabilities
🔀 Federate Elasticsearch data with other sources

and more! Register➡️ https://t.co/dKOlYuSVaL

[Praveen @praveenweb]

Similar to Google trends, I'm interested to see the most commonly used phrases, sentences, questions, or topics in any of the chat applications powered by LLMs (GPT, Claude, etc). Not the output produced, but more about the input from the user.

Separately, there is more of a revenue opportunity for OpenAI and Anthropic as well to collaborate with brands. For example: if a brand is mentioned in any of the chat conversations, OpenAI could show analytics data to verified brands. This is similar to Google Analytics / Search Console.

Probably more relevant now with SearchGPT? Of course, the data needs to be anonymized.

[Brandon 🚀 Flightcontrol @flybayer]

🔥 ANNOUNCING @FLIGHTCONTROL V3 🔥

Unveiling our true identity as a Platform-as-a-Service

There are many PaaS, but none as reliable, customizable, and affordable at scale as Flightcontrol ✨

🎒 We abhor one-size-fits-all thinking and overly prescriptive solutions. Our product enables you to take responsibility for critical decisions instead of us ushering you into The One True Way™️

🤝 We treat our customers right. And not just right, we go above and beyond to make you as successful as possible.

🌱 We thrive on sustainability. We are now long-term sustainable and do not need any more funding to continue thriving and growing!

🔎 We operate with utmost fairness and transparency. We'll never force you into an enterprise plan based on some unexpected usage that was too hard for you to estimate.

V3 is a HUGE improvement for us:

Deploy your apps with:

🏎️ 2-6x faster builds than every other PaaS with brand new build system!!

🤑 50-75% cheaper compute at scale than other PaaS with new ECS+EC2 support (unlocks ability to save even more with Reserved Instance pricing)

👾 now serving AI & video workloads with GPU support

🫣 new private web services not exposed to internet

✈️ scale further with server sizes up to 24,000 GB of RAM

🗿near perfect reliability because we use native AWS infra w/o custom layers

🧘‍♀️ utmost flexibility because it deploys to your AWS where you can customize or integrate anything, should the need arise

💰 Use your AWS credits to pay for infra AND Flightcontrol, because we are now on AWS marketplace!

How is all this possible?

You get all this in exchange for taking 1 minute to create & connect an AWS account (really, it’s not hard!)

Our bring-your-own AWS model enables us to leap-frog traditional PaaS on almost every metric. We don’t have to reinvent the wheel like other PaaS. We let AWS do what it’s best at, and that frees us to invest everything in DX and features.

Does AWS strike fear in your heart?

Fear not, we guarantee support for ALL the AWS stuff we manage. We handle it all for you, like any other PaaS. And with strict compute controls, you don't have to worry about run away costs.

We don't have millions of free users, so we can focus on providing a high level of support to our paying customers.

What to look forward to?

Lambda y'all!!! Our next big project is Lambda support for serverless web deployments and event processing. Soon you'll be able to manage and deploy both servers and serverless from a single platform ✨

Learn more at https://t.co/K78knFZ0yZ

P.S. Flightcontrol is not for you if you want totally free or ultra low cost infra. We are currently optimized for scalable production applications.