Missed it again....

Nobody cares.

Missed another standup, this time nobody bothered to pull you in ...

Deep in the flow of building something real, time doesn't pass -- it disappears.

Stop Playing Mute Tag. I'm Building the Fix.

You are mid-procedure, on a roof, in a deposition, at your kid's soccer game. You fire off a text to your admin with 4 things that need to happen. How many actually get done? I broke down 7 real scenarios — and what changes when the admin never forgets a step. 

Jessica — Your New AI Chief of Staff

Investigation leveraging 'embedded Claude' the why and the where.

Claude in a box gains Vision!

Bringing Vision to Claude

Claude in a box-4 - showing off a 'Claude Vision' reel. This is what Claude in a box or on a server can now handle Vision tasks that weren't there last week. Ask me how I built this skill and how you can use it in your work? Happy to have a conversation, you can reach me at https://petergits.com If you want to read the post prior to this one (Claude in a box-3), https://lnkd.in/etbwBA8D #ClaudeAI #IntelligentAutomation #RPA #AIAgents #SelfHealingAutomation #Playwright #BrowserAutomation #AnthropicAI #AIIntegration #DeveloperCommunity #TechInnovation #FutureProofing #AutomationEngineering

Adding hearing to Claude's skills vs CAPTCHA

# Building ChatCal.ai - Getting Calendar Booking to Actually Work **Subject**: Building an AI calendar assistant from scratch - OAuth, Claude Sonnet 4, and Google Calendar --- I wanted to let people book meetings with me without showing them my entire calendar. Sounds simple, right? Turns out there's a lot between "idea" and "working prototype." Then you get down to the real bit, Speech-to-Speech. ## What I Built A conversational AI that handles meeting requests like "book 30 minutes with Peter tomorrow at 2pm" and creates actual Google Calendar events with Meet links. Built with Python FastAPI, Claude Sonnet 4 from Anthropic, and the Google Calendar API, all running in Docker containers. ## The Tools I picked Claude Sonnet 4 because it's good at extracting information from messy human language. Someone says "this is Betty, call me at 630-800-9000" and Claude figures out that's a name and phone number. The Google Calendar API handles the scheduling, and FastAPI serves it all up with a simple chat interface. ## What Actually Took Time Google OAuth2 was my first weekend. Setting up redirect URIs, configuring scopes, understanding the consent screen - the docs are thorough but there's a lot of trial and error. I spent more time on OAuth than building the chat interface. Getting email invitations working with .ics calendar attachments took another evening. SMTP configuration, email templates, making sure both people get notifications - small details that matter when someone's actually using this. ## Early Challenges Claude sometimes got creative with extracting contact info. "Book a meeting with Sarah" would work, but "Sarah wants to chat next week" might miss the name entirely. I added better prompting and examples to make it more reliable. Google Meet link generation worked once I found the right API parameters. The Calendar API has a `conferenceData` field that creates real Meet links, not fake ones. ## Was It Worth the Free Time? I learned way more about OAuth flows than I expected. And debugging why calendar invitations weren't showing up taught me about email headers and RFC 5545. Not glamorous, but useful. Next: Making it fast enough for real use. --- 📅 **I'm sharing this journey in 4 parts. Should I post the rest:** - a) Daily (bite-sized updates) - b) Weekly (time to digest) **Vote below!** 👇 --- **Architecture diagram**: See chatcal-ai-design-document.html for the complete flow **Stack**: Python, FastAPI, Claude Sonnet 4 (Anthropic), Google Calendar API, Docker, SMTP **Time invested**: ~3 weekends (OAuth setup, email integration, Claude prompting) **Note**: Full source code freely available on GitHub (details in Phase 2 post) #AI #Python #GoogleCalendar #OAuth #BuildingInPublic #SideProject

# Making ChatCal.ai Fast - The Groq Migration and Why Response Time Matters **Subject**: Switched from Claude to Groq for 10x speed - learned why response time matters more than I thought (This is post 2 of 4 in a series) --- ChatCal.ai worked, but waiting 2-3 seconds for every response felt slow. When someone's trying to book a meeting, that delay adds up. Time to make it faster. ## The Performance Problem Claude Sonnet 4 is great at reasoning, but each response took 2-3 seconds. For a back-and-forth conversation - name, time, confirmation - that's 10+ seconds of waiting. Not terrible, but not snappy either. ## Why I Switched to Groq Groq's Llama-3.1-8b-instant gave me ~500ms responses. That's the difference between "thinking..." and "instant." Plus the cost dropped from $3/million tokens to $0.60/million. I'm paying for this myself, so that matters. The trade-off? Claude's reasoning is slightly better for complex requests. But for scheduling meetings, Llama-3.1 handles it fine. I'll take the 5x speed improvement. ## The Memory Problem Early versions forgot context immediately. You'd say "I'm Betty" and two messages later it would ask your name again. Frustrating for users, obvious I needed conversation persistence. I added ChatMemoryBuffer to keep conversation history. Now it remembers names, emails, and what you talked about. The system even retains info for future conversations - if Betty books again next week, it already knows her email. ## What Failed First I tried keeping full conversation history. Hit token limits fast. A long conversation would blow through my context window and start dropping messages. The fix: Smart context windowing. Keep critical information (name, email, phone), summarize or drop the rest. Not perfect, but works. ## Other Improvements Added custom meeting IDs in MMDD-HHMM-DURm format (like "0731-1400-60m" for July 31, 2pm, 60 minutes). Makes it easy to reference meetings for cancellations. Built intelligent cancellation matching - say "cancel my Friday 2pm meeting" and it finds the right one without needing the ID. Works by matching user info + date/time. Switched to HTML-formatted responses for better readability. Styled confirmations, color coding, professional layout. ## The Saturday I Lost Debugging why context kept disappearing took an entire Saturday. Turned out I was reinitializing the memory buffer on every request. Moved it to session state and everything clicked. ## Was It Worth It? The speed difference is noticeable. Conversations feel natural now instead of waiting for responses. And saving 80% on API costs meant I could actually afford to keep this running. Next: Adding voice input so people don't have to type. --- **Metrics**: 2-3s → 500ms response time, $3/M → $0.60/M tokens, conversation memory added **Stack**: Groq Llama-3.1-8b-instant, ChatMemoryBuffer, Python FastAPI, Redis session storage #AI #Performance #Groq #LLM #Optimization #ProductDevelopment

# Adding Voice Input - When HuggingFace Spaces Fights Back **Subject**: Built WebRTC speech-to-text on free GPU - FastAPI+Gradio don't play nice on HuggingFace --- Typing "book a meeting tomorrow at 2pm" is fine. Saying it should be better. Time to add voice input using free GPU resources on HF. ## The Plan vs Reality I wanted WebRTC audio capture in the browser, streaming to Whisper STT running on HuggingFace Spaces (free H200 GPU). Seemed straightforward. It wasn't. ## Week 1: The FastAPI Trap Tried combining FastAPI for WebSocket endpoints with Gradio for the ML interface. HuggingFace Spaces hated this. Port conflicts, mount failures, 500 errors with no useful logs. The platform is optimized for Gradio interfaces, not hybrid FastAPI apps. ## Week 2: WebSocket Dreams Die Attempted pure WebSocket implementations for real-time audio streaming. HuggingFace Spaces only reliably supports HTTP/HTTPS traffic. Their proxy layer doesn't handle WebSocket connections in containers. Tried Socket.IO, tried different ports, tried bypassing their proxy. Nothing worked consistently. WebSockets are a no-go on HF using containers. ## Week 3: The Breakthrough Switched to HTTP-based chunking. Browser captures audio via WebRTC, but sends it in chunks over regular HTTP. Not true real-time streaming, but good enough. Gradio's client expects files wrapped with `handle_file()` function. Documentation didn't make this obvious. ## What Actually Works Browser captures 16kHz mono audio using navigator.mediaDevices. Voice Activity Detection (energy-based filtering) identifies when someone's actually speaking. Only chunks with voice activity get sent to avoid wasting GPU time. Audio goes to Gradio client wrapped in proper format, hits Whisper base model on H200 GPU, returns transcription in 6-8 seconds. Not real-time, but usable. ## The Cost Win Free H200 GPU vs paying $0.006/second for commercial STT APIs. For a side project, that's the difference between "affordable" and "not happening." ## What I Built for Testing Trialed MCP Voice Service. Voice-like sine waves at 300-500Hz with proper energy levels. ## The Free-Time Reality Four days of evenings debugging why FastAPI wouldn't mount, why WebSockets dropped, why file uploads failed validation. Could've just paid for a commercial STT API and been done. But I learned HuggingFace Spaces limits, Gradio's file handling quirks, and WebRTC. Knowledge worth having for future projects. ## Still Not Perfect 6-8 second latency is higher than I wanted. Commercial solutions hit 2 seconds. But for free GPU compute, I'll take it. Next: Adding voice output to complete the speech-to-speech loop. --- **Tech**: WebRTC, Whisper base (OpenAI), HuggingFace Spaces (H200 GPU), Gradio client, Voice Activity Detection **Time**: 4 evenings debugging, ~20 hours total **Code**: Full implementation available on GitHub: https://lnkd.in/gNvSDeBu #WebRTC #SpeechToText #HuggingFace #AI #VoiceAI #BuildingInPublic #OpenSource

This is part 4 of 4 on how to spin up your own voice calendaring system for almost nothing (no cost, super low cost).

Nobody cares.

I was in two Zoom calls at the same time yesterday. Not sort of. Not "had someone cover for me." Both calls. Same time. I was represented in both rooms. Here's what actually happened. 2pm. Board prep call I'd been building toward for two weeks. Also 2pm — a client check-in that had been rescheduled three times and had a contract question sitting inside it that needed my answer before end of day. I didn't cancel either one. I didn't send an apologetic conflict message. I sent Jessica to both. Not a note-taker. Not a bot that records and transcribes. My delegate. She dialed into both Zoom calls. Introduced herself as my representative. Told each room I'd be joining when I was needed. Nobody was confused. Nobody felt like they were getting a lesser version of the meeting. They knew exactly who they were talking to and why she was there. And then she got to work. On the client call, she listened and summarized in real time — a running brief landing in my phone while I was in the other room. When the client raised the contract amendment, Jessica didn't wait. A clean message hit my inbox: "Client is requesting approval on [amendment]. Reply yes to proceed or no to hold." I replied in twenty seconds without leaving my board call. On the board prep, when they hit the agenda item that needed my voice, Jessica didn't take a note and move on. She dialed me in. I joined, gave my input, and stepped back out. She held the room while I was gone. She brought me back when I was needed. I was in both meetings. Jessica isn't a clone. She's a delegate. There's a difference worth understanding. A clone pretends to be you. A delegate represents you — keeps things moving, keeps you in the loop, sends you exactly what needs your decision, and brings you in only when you actually need to be there. Everything else? Handled. The double-book used to mean losing one of the two meetings every single time. A conflict message. An apology. A reschedule that takes another two weeks. Now it means nothing. This is what I'm building. Right now. If this is something you'd actually use — hit like and I'll personally ping you when we open the beta. I want the first people in to be the ones who felt this post in their chest. Link in the first comment 👇

Real estate agent Marcus is driving to a closing. Three properties need last-minute changes. He voice-texts his assistant: "Move Elm Street open house from Saturday 1-3 to Sunday 2-4. Lockbox code is 7742 — only give it to agents through ShowingTime, do NOT put it in the MLS notes. Cancel Oak Ave entirely, pull the social posts I scheduled. And call the photographer for Pine — tell her it is a 4-bed not 3-bed, the loft counts." Three properties. Six action items. One compliance-sensitive lockbox instruction. What usually happens: The time changes but not the day — still says Saturday. The lockbox code ends up in the MLS remarks out of habit. Oak Ave gets canceled but the social posts auto-publish Saturday morning for a listing that no longer has an open house. The photographer never gets the bedroom correction. What Jessica did: "Elm Street — moved to SUNDAY 2-4 (different day, confirmed). Lockbox 7742 restricted to ShowingTime only, NOT in MLS. Oak Ave — canceled, 3 scheduled social posts pulled. Pine — photographer confirmed for Friday, corrected to 4-bed." Three properties. Zero mistakes. No compliance violations. Link in comments.👇 #RealEstate #OpenHouse #AIAdmin #JessicaAI #RealtorLife

Monday morning. The dentist is running 20 minutes late. She texts her front desk: "Move my 9:00 to 9:20. Change it from a cleaning to cleaning + exam — that is 60 minutes not 30. Check if that pushes into the 9:40. If it does, call the 9:40 and offer 10:30. If they can't do 10:30, cut the 9:00 back to cleaning only." There is math in this text. There is conditional logic. And there is a fallback plan. Her admin moved the appointment and changed the type. Did not do the math. 9:20 plus 60 minutes is 10:20. That overlaps the 9:40. Both patients showed up. Chaos in the waiting room. Jessica did the math before touching the calendar: "9:20 + 60 min = 10:20 — that DOES OVERLAP the 9:40 (Thompson). Calling Thompson now to offer 10:30." Thompson said yes. Updated schedule: - 9:20 — Williams, cleaning + exam (60 min) - 10:30 — Thompson, cleaning (30 min) No overlaps. No surprises. No waiting room collisions. Your admin should not need to be a mathematician. Jessica already is one. Link in comments.👇 #DentalPractice #Scheduling #AIReceptionist #JessicaAI #SmallBusiness

Saturday afternoon. Your kid's soccer game. Three clients reach out in an hour. One has a tax deadline with a 25% penalty if she misses it. One wants a routine paperwork change. One is a brand-new lead asking about your services. You text your admin: "Patterson is urgent — call her back TODAY. Baker needs 30 minutes, not 15, schedule next week. Garcia is a new lead — get her info but do NOT discuss any products or give financial advice." Three contacts. Three different priority levels. One compliance guardrail. A human admin schedules all three for "next week." Books Baker for the default 15 minutes. Chats with Garcia about the market because she asked. Now you have a missed deadline, a short appointment, and a compliance issue. Jessica: Called Patterson back within 5 minutes. Monday 8:30 AM confirmed. Baker scheduled Wednesday, 30-minute slot. Garcia — lead captured, referral source noted, zero product discussion. "Enjoy the game." Your weekends should be yours. Jessica keeps working. Link in comments. Link in comments.👇 #FinancialAdvisor #WorkLifeBalance #AIAdmin #Compliance #JessicaAI

Mike owns 6 HVAC trucks. He is on a roof when his biggest client calls the office in a panic — AC out in a 40-unit building. In July. He voice-texts his admin: "Pull Danny off Johnson, send him to Parkview, it's an emergency. Reschedule Johnson to tomorrow. Tell Danny he needs the 410A tank. Get the building super's cell from Parkview. Bill it to the PM account not the tenant." That is 6 action items across 4 different people. His old admin? Called Danny but forgot the 410A. Called Johnson but did not reschedule. Never got the super's number. Danny showed up on a roof he could not access with the wrong refrigerant. Jessica? Parsed all 6 items. Called Parkview first (because Danny needed the info before leaving). Texted Danny a complete briefing. Rescheduled Johnson with an apology and a specific time. Flagged billing correctly. 40 minutes. Every item done. Every person contacted. Zero dropped balls. That is the difference between "handled" and actually handled. Free 30-day trial. Link in comments.👇 #HVAC #SmallBusiness #AIAdmin #FieldService #JessicaAI

Dr. Patel is mid-root canal. She pulls off a glove and fires off this text to her front desk:  "Move Mrs. Chen from Thursday to Tuesday AM. Cancel Rivera tomorrow. Ask Chen if her insurance changed. Oh and check if Tuesday has a hygienist slot before you confirm."                                                   That is 4 tasks, 1 dependency, and zero chance her admin catches all of it between the walk-in at the counter and the insurance call on hold.                                                   Here is what Jessica texted back:                                    "Got it. Here is my plan:                                     1. Check Tuesday AM for a hygienist slot            2. If open, call Mrs. Chen to reschedule  3. During that call, ask about her insurance  4. Cancel Rivera's Wednesday cleaning                                                    I moved #1 to the top because the reschedule depends on it. Will text you when each one is done."                                                                                               20 minutes later, all 4 done. Insurance flagged. No overlaps. No follow-up needed.                                        Your front desk is doing their best. Jessica never forgets a step.                                                100 free seats this month. Link in comments.  👇                                                         #AIReceptionist #DentalPractice #MissedCalls #JessicaAI #SmallBusiness 

"He's free — the company pays for him." Cool. So does your competitor. And the client who just hung up. You're sharing an admin. That sounds like a perk until your client calls while you're on another call. Now they're on hold. Or worse — voicemail. Or worse — they move on. Let's be honest about what a shared admin actually does for you: Takes a message. Books an appointment. Says "I'll have him call you back." You could do that yourself. What he can't do is answer the way you answer. He doesn't know your process, your pricing, your language, your nuance. He's not you. He's a time-share — and right now, someone else is using your week. And at 6pm? He's gone. Your clients aren't. You're not paying his salary — you're right about that. But you're paying in missed calls, lost impressions, and customers who felt like an afterthought. The business owners winning right now aren't sharing. They have something built around them — that knows how they speak, what they offer, and never puts a client on hold. Stop renting a slice of someone else's assistant. Get one that's actually yours. Link in the first comment 👇 #LinkedInTips #SmallBusiness #Entrepreneurship #BusinessGrowth #CustomerExperience #Productivity#Executive

You hired 1/5 of a person. Which means right now, while you're reading this, she might be answering someone else's phone. Not because she doesn't care. Because that's the math. She works for four other executives. Four other businesses that all believe, reasonably, that they're the priority. And she can't answer two calls at once. Nobody can. So when your call comes in while she's handling theirs — it goes to voicemail. Every time. That's not a people problem. That's a physics problem. And it has been quietly costing you first impressions, qualified leads, and booked appointments every single day. I built Jessica. She answers your phone. Always. She is never on another call. She is never handling a different executive's schedule. She is never eating lunch or finishing a sentence or away from her desk. Your calls never go to voicemail. Not once. Not ever. But she's more than a call-handler. She's your shadow. She knows your calendar. She knows where you need to be. She'll remind you before the meeting, not five minutes after it started. You can talk to her. You can text her. She texts you back. And here's the part that's different from every other setup you've seen: When she connects you into a call — she doesn't hand the caller off to dead air and disappear. She stays on the line. Keeps the conversation warm, live, and moving right up until the moment you join. No hold music. No sixty-second silence while your caller mentally checks out. No "please hold while I transfer you." You walk into a call that's already alive. The impression your caller leaves with? You run a tight operation. You have real support. You're serious. Because you are. You just didn't have the tool until now. 100 free seats. 30 days. No credit card. Your calls start getting answered today. Link in the first comment 👇

I do all my best technical work in the morning. Always have. A few months ago I added intermittent fasting to the mix. Last meal by 6pm, nothing until noon the next day. If you know, you know — the mental clarity in that fasted state is something else. Focus is sharper. Thoughts are cleaner. You get into a zone that's hard to describe until you've been in it. It works great. As long as you don't schedule meetings. Well — it works great as long as you turn your phone off. And like most people, I can't do that. Telemarketers I can handle. They tend to show up around 5–6pm and their area code usually gives them away. But I have clients, leads, vendors, people who genuinely need to reach me — and those calls I have to answer. Even at 8am. Even when I'm finally, actually, in the zone. So I built Jessica. She's an AI admin who knows my schedule. She answers every call. Books appointments directly into my calendar. Screens out the telemarketers so they never reach me. And when something is genuinely urgent — when someone actually needs to speak to me right now — she patches them through. That's it. That's the whole thing. I got my mornings back. I'm utilizing the fast the way it was meant to be used. The focus is there. The clarity is there. I'm actually ahead in my work this week for the first time in longer than I want to admit. And honestly? It's not so much the fasting. It's Jessica. --- I'm opening 100 free seats this month for business owners and professionals who want the same thing. 30 days, no credit card, no contracts. You control when calls reach her. We connect her to your calendar. She handles everything else. We don't know how long unlimited calls stays in the free tier — so if you've ever wanted to take conversational AI for a real spin, this is your moment. Take Jessica for a ride. Link in the first comment 👇 ---

Production AI vs. Conversational AI. Almost nobody knows the difference. Everyone in this feed is claiming victory with AI. And most of them are right — about yesterday's AI. Production AI. Workflow AI. The kind that automates a task you already defined, inside a system you already built, with rules you already wrote. That's not wrong. It's just not *this.* Conversational AI is a different league entirely — and almost nobody in business is actually using it yet. Not really. Not in the raw, live, unrehearsed way that shows you what's possible right now. Today. I built Jessica. She answers your phone. Holds a real conversation with whoever calls. Qualifies them. Handles their questions. Books them directly into your calendar, exactly the way you've told her to. Adjusts. Reschedules. Manages the vendor who keeps calling at the worst time. Tells the person who calls three times a week the same question exactly what they need to know. And she does it while you're doing something else. Here's the part I want you to actually feel: You wake up tomorrow morning. Coffee. No phone ringing. No "quick question" interrupting your first clear thought of the day. No scheduling back-and-forth eating the hour you needed to actually think. Jessica already handled the 7am call. Already booked the 10am. Already told the vendor you'll call back Thursday. Your morning is yours. I can hear the paint dry now — and I mean that as the highest possible compliment. That stillness, that focus, that uninterrupted stretch of time where you actually get your work done? That's what conversational AI *actually* delivers. Not a prettier dashboard. Not a faster report. Your mornings back. This is not the AI of six months ago. It keeps getting better, faster than most people realize — and if you haven't put your hands on it yet, you are genuinely missing it. Not "falling behind the curve" missing it. *Missing it* missing it. Here's your moment to dip your toe in. 100 free seats. 30 days. No credit card. You control when your calls go to Jessica and how — simple call forwarding, two minutes of setup. We connect her to your calendar. She handles the rest. We don't know how long we can keep unlimited calls in the free tier. Right now that's what it is. Take Jessica for a ride. Even if you're just curious. Especially if you're just curious. The lake is warm. The water's right there. 100 seats open every month — this month's are going fast. Miss it and you're waiting until next month to start. Link in the first comment 👇