Automating complex financial systems.
New York, NY
Joined November 2022
- Tweets5.2K
- Following178
- Followers822
- Likes6.8K
Multimodal retweeted
Most agentic AI demos are easy. Production agents in credit union and community bank lending workflows are hard.
They need evals that test the whole workflow, not just the final answer. Regulation. Auditability. Document understanding. Exception handling. And a clear sense of when to stop and hand off to a human.
That last part is the whole game. A 95% extraction score still fails if the agent can't tell which 5% to flag for a loan officer.
So we build the boring parts: logs, evals, permissions, rollbacks, human review queues.
The boring parts are the product.
Building this at @MultimodalAI. Following along here.
ALT Regulation is not the ceiling. It's the floor.
Multimodal retweeted
The finance ops problem nobody talks about:
Your best analyst isn't slow because they're bad.
They're slow because they spend 60% of their day pulling data from systems that don't talk to each other, formatting it into templates, and waiting for approval chains that haven't changed since 2004.
The bottleneck isn't intelligence. It's plumbing.
That's what we're fixing. @MultimodalAI
Multimodal retweeted
Anthropic CEO just said Ai will wipe out 50% of lawyers, consultants, and finance professionals within 12 months: xitter.cf/theUMreal/status/20269…
Readers added context they thought people might want to know
As is evident in the video, Amodei said that 50% of all *entry level* white color jobs could be replaced, and his timeline is 1-5 years, not strictly 12 months. Further confirmation from other news sources:
axios.com/2025/05/28/ai-…
Multimodal retweeted
The current wave of AI is human-to-agent: you ask, it answers.
The next wave is agent-to-agent: one agent completes a step, hands off to another, and the workflow keeps moving.
That’s the trend I’m most excited about: orchestration.
—>The reason is simple, business workflows don’t happen in one system.
They span tools, teams, vendors, and data sources. And most of the friction is coordination.
The real unlock is when agents can synchronize across that complexity: passing context, triggering actions, and collaborating even when they’re not built by the same company.
That’s what turns “AI features” into “AI operations.”
If you could automate one cross-tool handoff end-to-end, what would you pick?
#AiFuture #Aitrend
Multimodal retweeted
Someone recently asked me, if I had to explain what I did with @MultimodalAI to a smart 12 year old, how would I explain it to him?
I would tell him that, We’re building AI coworkers that help humans do their jobs better.
They take care of boring, repetitive admin work,
so people can focus on thinking, helping customers, and making better decisions.
When humans work better, customers get better experiences. That’s the whole point.
And I personally think this clarity is great especially for people out the Agentic AI space that doesn’t understand what it is.
#simplifyingAgenticAi #Cartoons
Multimodal retweeted
Appreciate your feedback!
Glad to hear my Bard tutorials are hitting the mark.
Making marketing less hard, one video at a time!
Multimodal retweeted
Hey @maraza24!
We're loving your app! 😍 langchain-chat-pandas.stream…
Would you be interested in writing a blog post about it on our Streamlit blog?
Multimodal retweeted
Runway Gen-2 Image-to-Video is a game changer 🤯
You can craft a music video or short film draft from Midjourney still images, combined with music/voice in record time.
Here is how I created this music video and 11 INSANE examples, Sound ON
(THREAD 🧵) 1/13
Multimodal retweeted
AI is getting complicated with each passing day.
5 months back, all we knew was "ChatGPT."
that was simple.
Now there're so many AI terms for beginners to understand...
like:
• LLM
• AGI
• T2I
• NLP etc
I thought let's simplify all these major terms of AI in 1 Tweet.
So here it is:
• LLM (Large Language Model ): Very big math equation that is very good at predicting the next word.
What are the top Large Language Models do we have right now?
• ChatGPT by OpenAI
• Claude 2 by Anthropic
• LLaMA By Meta
• Generative AI: A category of AI that can "produce" content like text/image.
• Natural Language Processing (NLP): NLP is a part of AI that helps computers understand, interpret, and generate human languages, like turning spoken words into text or answering questions in a chatbot.
• Multi-modal AI: A type of AI that can process different types of inputs such as text, speech, images, and videos.
• Prompts: Messages you send to AI (for ex: ChatGPT or MidJourney)
• Input: What you provide to the AI, usually in the form of prompt or question.
• Output: The AI's response to your input.
• Prompt Engineering: The "Science" of sending the exact right message to ChatGPT to get the output you want.
• Fine-tuning: The process of training a pre-trained model for a specific task. For ex: summarization, text generation etc.
• Token: The smallest piece of text that the AI can read. Could be a letter or a word.
• Embeddings: Similar to tokens, embeddings are another way to encode text as numbers. However, there are a few differences, embeddings can represent words, sentences or whole passages of text.
• AI Agent: AI system that can achieve the taget goal by creating & implementing tasks, by itself.
- AgentGPT
- AutoGPT
- BabyAGI
In technical terms, most AI agents are developed through chain-of-thought prompting technique.
• text-to-image Model (T2I): An AI model that produce images from text prompts.
- Stable Diffusion
- MidJourney
- DALLE
• text-to-text Model (T2T): An AI model that is trained to produce text form content like:
- Email copy
- Ad copy
- blogs etc
• text-to-video (T2V): An AI model that is trained to produce video form content from prompts or scripts.
- Synthesia
- Runway
- Fliki
• text-to-audio (T2A): An AI model that is trained to produce audio form content from trained data, or description.
- Descript
- Murf AI
- Play. ht
• Knowledge base chatbot: A chatbot that is trained on your data. For ex: your website data, PDFs, reports etc.
- Chatsimple
- Botpress
- Dante AI
*** Popular AI Dev. programs
• Code Interpreter: It is an official plugin of ChatGPT. Code interpreter allows ChatGPT to analyze data, create charts, solve math problems and edit files, among other uses.
• Chat Plugins: ChatGPT plugins are add-ons you can pair with the AI chatbot (ChatGPT) to extend its capabilities like researching internet, creating videos, charts, etc from within ChatGPT interface.
• Langchain: LangChain is an open-source library that provides developers with the tools to build applications (apps) powered by large language models (LLMs).
• Pinecone: ChatGPT can't save your chats in its memory. Pinecone provides the structure of long-term memory for AI.
If you want to build your future, educate yourself on latest AI development.
Then practice & practice. I guarantee you, you'll find your life-changing moment with AI.
I am posting every day about AI and Automations on Twitter, follow me @cj_zZZz for actionable AI content that can make you money.
Retweet this tweet to share this knowledge with your audience.
Sign up for 'The AI Startup' newsletter for exclusive content on AI.
Cheers.
Multimodal retweeted
It's been 3 days since I launched ThreadWiz beta.
I received more feedback than I first anticipated, thanks to everyone who contributed!
And in full #buildinpublic disclosure, here are the key lessons I've learned so far. 👇
Great lessons for anyone looking to build products.
You never know how people will respond until you launch 🚀
You can't predict how people will react to what you've created until it's out there in the open.
It is a well known fact that the market is the only real feedback you can have.
When your product is just an idea, you have to make many assumptions about what people will like or dislike.
You may ask some people what do they think about your new idea... You can make a couple of tweets to gauge reactions...
But nothing compares to actually launching it.
You see, talk is cheap.
Asking people what they think of your app or product idea is one thing.
But having them commit their own time and money to use it is an entirely different game.
Not surprisingly, that's the only way you can know if you are building something useful or not.
So, if you have a product or app idea, just launch it and start gathering feedback.
Any feedback you get before people have to invest their own time and money, is not accurate.
And if you're avoiding launch day by seeking indirect feedback, you're probably just procrastinating.
Sh🤬 happens.
The first group of test users encountered bugs that I had never seen before.
Bugs related to the API, the app flow, the buttons... You name it.
Murphy's law exists for a reason.
Things that can go wrong will go wrong, and you can't predict exactly what until it happens.
Again, just launch.
Now, the reason I stress the "just launch" motto so much is that, based my conversations with other creators, most of them are battling with perfectionism.
They are constantly seeking indirect feedback instead of releasing their products out in the open and facing reality.
I know full well that getting over perfectionism can be a challenge.
But if you shift your mindset from delivering the perfect product to releasing a good enough MVP, you'll relieve a lot of pressure and start taking more risks.
Did I build something useful? 🤔
Now, the most pressing question I had in my mind was: "Am I even building something useful?"
Of course, it's still way too early to answer this question.
But the good news is that I already have some returning users and I made some sales, which suggests that my app has potential.
So for now, I will continue building the app based on the feedback I have received.
Stay tuned. 👀
PS: Wrote this ~450 words post in 15 minutes using the draft editor.
Honestly curious to know how long you would take to write and publish a post with a similar length.
Multimodal retweeted
Don't pay $20 for ChatGPT or any other ChatGPT tool
Introducing Merlin – the one & only extension you'll ever need.
Here're 10 features that make Merlin the only ChatGPT extension you need:
Multimodal retweeted
Coding used to be a marathon. Now it's more like a sprint.
With AI, apps are whipped up in minutes, no code necessary.
Here is how I create an app with AI in just a few clicks:🧵
Multimodal retweeted
Thanks for your lovely wishes and blessings.
Here are our 7 winners for this Giveaway:
1. @MultimodalAI
2.@AIMktgWhiz
3. @robertjackson
4. @sampson_branson
5. @SirStegi
6. @JvShah124
7. @pranaysuyash
Congratulations to all winners.
Next Giveaway at 35K. Stay tuned.
Thanks for constant support 🙏
Multimodal retweeted
SutraGPT Update:
Three days ago, I asked you guys to help me build SutraGPT.
And boi, I got an overwhelming response. 🤯
Thank you so much everyone for submitting your valuable questions. (link in comments)
If you have any more queries, please comment below.
The front end is ready as you can see in the below picture.
Lmk how it looks. ⬇️
Now I need to train the Chatbot.
So here's what I am gonna do:
• Build an FAQ document answering the questions you guys have asked
• Upload the doc on @chatsimple_
• Train the chatbot
• Launch for alpha testing
Why am I doing this? Cuz I can and it's so fun. ^_^
LFG!!
Multimodal retweeted
Here's my biggest hack for for creating content with AI.
This hack singlehandedly reduces the time it takes me to produce unique, engaging, and original content by 80%.
Easily skyrocketing my content output.
So much so, that I can write content practically as fast as I can write my thoughts down.
On top of that, this writing hack is the only way I've found for avoiding the perils of AI-generated content. (That is, robotic sounding and deathly boring content)
Here's how it works, feel free to bookmark it for later reference:
1. Write as quickly as you can ✍️
I've found that the best way to infuse your unique ideas into the content is to write them yourself.
Don't worry about formatting, grammar, or even punctuation.
Just write everything down as quickly as you can, even using placeholders when you're unsure of the right word.
For instance, take a look at the video attached in the post...
Pay attention to how badly written is my first draft.
I use lots of "??" for when I forget which word to use and the entire content barely makes sense.
Again, the one and only goal at this step is to get your thoughts out of your head and onto the screen as fast as possible.
Don't worry about editing just yet, simply write down your entire post in one go.
2. Let AI handle the editing 🤖🗒️
While GPT may not be the best writer, it's an excellent editor.
It can take your rough draft, no matter how nonsensical, and turn it into a polished piece ready for publishing.
I simply copy and paste my first draft, send it to GPT, and let it do its magic.
Again, see the gif attached.
3. Review everything once more 👀
This step is more of a sanity check. Maybe you thought of something else you'd like to add or change. So, read it, make minor changes, and move on to the next step.
4. Publish 🖥️🖱️
And now that everything is ready, go ahead and publish your work.
As you can see, this system reduces the editing process from 5+ rounds of back and forth editing to just one, significantly shortening your writing time.
This system helps me publish a lot of content quickly, and it's the only way I've found to use AI for writing while keeping high standards for your content.
In fact, this very post you are reading was entirely made using this process.
If I had gone through the usual route of telling AI to write it for me, this post would be completely uninformative.
Now, if you find this method useful, I have developed an app to help you write using this system.
The app is free, the link is in my bio.
Hopefully you can start using AI the right way and skyrocket the amount of content you produce with less time spent on it.
If you have any questions about generating high quality content with AI, feel free to ask them in the comments.
