4d • General discussion
AI Developer Accelerator — Coaching Call - May 26th
AI Developer Accelerator — Coaching Call - May 26
Meeting Purpose
Key Takeaways
- Prakrit's AI learning app prototype is stuck. The solution is to use ShipKit's "chat template" for sequential/parallel LLM calls, avoiding RAG and using smaller, faster models (e.g., Gemini 3.5 Flash) for cost and latency.
- Scott's parking app generated $5k in its first weekend. This validates a replicable model for custom software: find a niche (e.g., parking, studio booking), build a solution, and scale it.
- Patrick released "Community Brain," a searchable RAG system for call notes. He also replaced OpenClaw with Hermes, an agent framework that uses Codex for coding and offers superior API key management.
- A new PDF processing workflow was shared: Convert PDFs to images, then use a multimodal LLM (e.g., Gemini) to convert images to structured Markdown. This preserves complex layouts (tables, flowcharts) that traditional OCR misses.
Topics
Prakrit's AI Learning App
- Problem: Prakrit's prototype for a spelling app stalled after switching to spec-driven development, which proved inefficient for a solo dev without a solid tech foundation.
- Solution: Use ShipKit's "chat template" for a sequential/parallel LLM workflow. Rationale: The app's core function is generating ~30 AI components from a single word input. Workflow: Decompose the word → use results to generate meaning, origin, etc. Model Selection: Use smaller, faster models (e.g., Gemini 3.5 Flash, GPT-4o Mini) for lower cost and latency. Output: Use structured outputs (e.g., JSON) for predictable data.
- Next Steps: Restart the ShipKit prep phase, feeding the existing spec and mockups to the AI for a structured development plan.
Bastian's Healthcare Chatbot
- Project: Pitched an appointment confirmation chatbot to a healthcare center.
- Outcome: The initial request expanded into a larger project: a state machine mapping the entire patient journey (appointment → payment) to fix fragmented funnels (web, email, phone).
- Business Strategy: Offered a partnership model instead of a flat fee. Terms: Pay-as-you-go for LLM costs, with a time-limited offer. Rationale: Positions the project as a long-term collaboration, not a one-off sale.
- Pricing: Pitched a high number (~$38k equivalent) to anchor value, justifying it by projecting a $200k/year revenue increase for the client.
Patrick's Community Brain & Hermes
- Community Brain: Released a pre-release version of a searchable RAG system for call notes. Tech: Docker, Open Web UI, GPT-4o Mini (or a local model like GPT-OSS-20b). Content: Fed by the Recap Flow, which generates weekly meeting summaries.
- Hermes Agent Framework: Replaced OpenClaw with Hermes. Rationale: Superior API key management and native support for using an OpenAI Codex subscription as a tool. Functionality: Aggressively uses sub-agents and tools (e.g., Firecrawl, OnShow) for tasks like research and coding. Example: Hermes forked its own repo and built an AI usage dashboard after a simple prompt.
Tom's Military Association App
- Project: Building a membership database for a military association (5,500 members).
- Opportunity: Two other associations have expressed interest, creating a potential paid product.
- PDF Processing: Goal: Ingest historical documents ("Chronicles") from 1926. Challenge: Traditional OCR (e.g., Doclink) struggles with complex layouts (tables, images). Recommended Workflow: Convert each PDF page to an image → use a multimodal LLM (e.g., Gemini Flash) to convert the image to Markdown. This preserves layout and improves RAG results.
Scott's Parking App & AI Tools
- Parking App: A custom app for a gym owner generated $5k in its first weekend. Tech: Twilio (SMS), Stripe (payments). Model: A replicable solution for property owners to capture 100% of parking revenue.
- AI News Digest: An automated newsletter summarizing AI news from 10-12 RSS feeds. Tech: Claude Haiku (summaries), Eleven Labs (audio), Resend (email). Opportunity: Monetize the newsletter using playbooks from successful newsletters like The Hustle.
- AI Video Generator Repo: A GitHub repo for generating videos from screenshots and text. Tech: Remotion, Eleven Labs, Higgsfield. Use Case: Automate creation of personalized video teasers for cold outreach.
Morgan's Projects & Hermes Setup
- Hermes Setup: Configuring Hermes on a VPS, following Patrick's recommendation.
- SignPy: A side project for creating AI-generated lobby display boards. Tech: Raspberry Pi (Node.js server, Chrome kiosk mode). Moat: AI generates beautiful slides from unstructured input.
- Heritage Plot: A cemetery management app. Challenge: PDF processing for plot documents, many with handwritten content. Solution: The PDF-to-image-to-Markdown workflow is a perfect fit.
- Google Workspace Automation: Automating email/PDF processing for a client. Challenge: High friction and disconnected APIs in the Google Workspace ecosystem.
Juan's AI Booth
- Project: An AI photo booth app for events.
- Tech: Stable Diffusion 4.5 (image-to-image), Wavespeed (inference), AWS CloudFront (asset delivery), Resend (email).
- Challenge: Cross-browser compatibility for QR code downloads (Safari vs. Chrome).
- Observability: A separate data web app provides AI FinOps to track costs per event and manage jobs.
- Recommendation: Use a log drain (e.g., Vercel → Google Cloud) to monitor logs and send alerts for errors.
Alex's Multi-Tenant Booking App
- Problem: The successful studio booking app needs to be multi-tenant to serve new clients.
- Recommendation: Build a new multi-tenant app from scratch. Rationale: Avoids breaking the existing client's service. Process: Build and test the new app with dummy organizations → port the existing client's data.
Next Steps
- Prakrit: Restart the ShipKit prep phase, using the existing spec and mockups to generate a development plan.
- Tom: Implement the PDF-to-image-to-Markdown workflow for the military association's historical documents.
- Scott: Share the AI Video Generator GitHub repo. Explore monetization strategies for the AI News Digest. Scale the parking app business by targeting other property owners.
- Morgan: Set up Hermes on a VPS. Implement the PDF-to-image-to-Markdown workflow for the Heritage Plot project.
- Juan: Compare Stable Diffusion 4.5 with the new GPT-2 image model. Implement a log drain for the AI Booth app to enable real-time error alerts.
- Alex: Begin designing and building the multi-tenant version of the studio booking app from scratch.
Action Items
- Re-run ShipKit prep w/ current specs + meeting transcript; start new project; implement sequential/parallel LLM calls w/ structured output + small models - WATCH (5 secs)
- Test new anonymizer for recap; publish recap w/ anonymized emails - WATCH (5 secs)
- Add Vercel log drain to Google Cloud w/ Slack/email alerts - WATCH (5 secs)
- Build multi-tenant studio booking app from scratch; migrate existing client after dummy-org smoke tests - WATCH (5 secs)
0
0 comments
AI Developer Accelerator — Coaching Call - May 26th
skool.com/ai-developer-accelerator
Master AI & software development to build apps and unlock new income streams. Transform ideas into profits. 💡➕🤖➕👨💻🟰💰
Leaderboard (30-day)
1
+6
2
+5
3
+2
4
+2
5
+2
Powered by