Experimenting with AI and email APIs

Gmail’s AI email assistant writes like a committee of lawyers designed it. Pete Koomen’s recent post Horseless Carriages explains why: developers control the AI prompts instead of users. In his post he argues that software developers should expose the prompts and the user should be able to control it. He inspired me to build my own. I want a system that’s fast, accounts for historical context, & runs locally (because I don’t want my emails to be sent to other servers), & accepts guidance from a locally running voice model. Here’s how it works: 1. I press the keyboard shortcut, F2. 2. I dictate key points of the email. 3. The program finds relevant emails to/from the person I’m writing. 4. The AI generates an email text using my tone, checks the grammar, ensures that proper spacing & paragraphs exist, & formats lists for readability. 5. It pastes the result back. Here are two examples : emailing a colleague, Andy (https://lnkd.in/gtjt3BPp), & a hypothetical founder (https://lnkd.in/gDwM4f22). Instead of generics, the system learns from my actual email history. It knows how I write to investors vs colleagues vs founders because it’s seen thousands of examples. The point isn’t that everyone will build their own email system. It’s that these principles will reshape software design. - Voice dictation feels like briefing an assistant, not programming a machine. - The context layer - that database of previous emails - becomes the most valuable component because it enables true personalization. - Local processing, voice control, & personalized training data could transform any application, not just email, because the software learns from my past uses We’re still in the horseless carriage era of AI applications. The breakthrough will come when software adapts to us instead of forcing us to adapt to it. Centered around a command line email client called Neomutt (https://neomutt.org/). The software hits LanceDB, a vector database with embedded emails & finds the ones that are the most relevant from the sender to match the tone. The code is here (https://lnkd.in/gZ-AaAWa).

Yesterday I had a call with 2 founders who wanted to make sales outreach more efficient without it sounding like AI slop. Together, we found a counterintuitive way to make that possible. Their sales team was spending hours crafting personalised emails. "Surely AI can help us?" seemed like the obvious question. But equally, they were painfully aware of being on the receiving end of AI generated sales emails and wanted to avoid that. So what could we do? - The starting point for any AI project is doing a process breakdown. So I asked them to walk me through their actual process. Step by step. Turns out they were spending 10 minutes checking data from multiple different places before even opening Gmail. The bottleneck wasn't writing the emails at all. It was gathering data. So instead of discussing how to use AI to write the emails, we discussed how to use it to solve the data issue. - The team was non-technical, but AI has opened up a world for non-technical people to start building technical solutions themselves. They had already experimented with asking Claude to write little bits of code (and asking Claude to teach them how to run it). Now is their opportunity to take that a step further. We discussed building a tool which brings all the relevant data together so the sales rep has everything they need to craft that perfect, personal email. We discussed how this tool could also help them better prioritise who was urgent to contact. And for bonus points, we also discussed using AI transcription tools like Superwhisper or Wispr Flow so the rep can 'speak' the email instead of typing. - With all this, I reckon we could get a 15 minute process down to a couple of mins PLUS potentially even making the emails BETTER by making all the context more accessible AND it's also way more fun for the sales team, who get to focus on building relationships instead of trawling spreadsheets. AND it's also better for the customer, saving them from yet more sales slop in their inbox win win win win! - Not good for my sales numbers though, as I gave them the DIY solution instead of pitching something complicated for Artanis to build for them ;) I think the best AI implementations are often invisible. They're solving the real, hidden inefficiencies that create bottlenecks, and enabling humans to do more of what they do best. Does this spark any idea for hidden bottlenecks AI could solve for you? -- 👋 Hi, I'm Laura Rosenberger I write about: • getting into AI if you're non-technical • building AI (that actually works) with Artanis • creating genuine impact in your business from AI

How I use Claude.ai to save time and hassle chasing submissions from businesses We’ve all had to wait on forecast information, or received incomplete submissions and then had to chase. The good news is that automating this task, which isn’t my favourite(!) is quick and easy using AI. To give an example, I’ve generated a script using a simple prompt in Claude.ai to read reports (in this case csv, word, pdf and powerpoint) and query unclear items or where an explanation is needed. This saves time and allows me to move onto other things. Here’s what the script does: Reads reports in CSV, Word (DOCX), PDF, and PowerPoint (PPTX) formats Automatically detects the file type and processes accordingly Analyzes the report content for unclear items or those needing explanation Generates a formatted HTML email with the findings Sends the email to relevant stakeholders Saves a copy of the email content as an HTML file for record-keeping This versatile automation can save hours of manual work across different document types and ensure that important issues are promptly addressed. And it’s not just useful for treasurers, it can be used by project managers, team leads, and anyone dealing with regular reporting processes using various file formats. Key features of the script: Uses pandas for data manipulation Implements file type detection and appropriate reading methods Utilizes libraries like python-docx, PyPDF2, and python-pptx for different file formats Generates clean, formatted HTML emails using Arial font Easily customizable for different report formats and email content I've included the full code in the comments below. Feel free to adapt it to your needs or reach out if you have any questions! This is just one example of the type of automation that we’re focussing on at Your Treasury - AI your way I’d encourage you to go to Claude.ai and try creating something similar yourself. #Python #Automation #ProjectManagement #DataAnalysis #EmailAutomation #DocumentProcessing ——————————————————- Example prompt Create a Python program where a user uploads a file and python then detects the file type (doc, excel, pdf, PowerPoint) and Python then reads a report and send out emails to relevant stakeholders if anything is unclear or needs more explanation. Produce example code to show this, including the report and a sample email which should be saved as an html file and formatted in Arial text. Prompt me if you have any questions