10d • General discussion
AI Developer Accelerator — Coaching Call - May 5th
AI Developer Accelerator — Coaching Call - May 05
Meeting Purpose
Key Takeaways
- Data Preprocessing is Key: Patrick's "Community Brain" RAG project shows that meticulous data preprocessing (costing ~$30) is the most valuable step, yielding an ~80% performance match to Opus with a local GPT-OSS model.
- New AI Dev Stack: Superpower is emerging as a critical framework that enables a shift from Claude to Codex. This move is driven by Claude's recent performance issues and Codex's new, more generous $100/mo tier.
- Hardware-Software Synergy: Morgan's Raspberry Pi kiosk engine provides a robust, offline-first display solution, which Juan plans to integrate with his "AI Booth Studio" for dynamic event content.
Topics
AI Developer Stack: Claude vs. Codex
- Problem: Developers are experiencing performance degradation with Claude, including increased token usage and slower response times, leading to frustration and stalled progress.
- Solution: Shifting to Codex, with Superpower as the common development harness. Superpower provides a consistent, high-quality development experience across platforms (GitHub Copilot, Claude Code, Codex). Codex's new $100/mo tier offers more generous usage limits than Claude's equivalent plan.
- Transition Strategy: Patrick: Plans to switch from Claude Max to Codex Max next month. Paul: Will adopt Codex and Superpower to resolve a stalled project. Bastian: Shared methods for remote access to a local Codex instance (QR code or SSH).
"Community Brain" RAG Project
- Goal: Build a RAG system to query 2.5 years of community call transcripts.
- Key Finding: Meticulous data preprocessing is the most valuable step, far outweighing the cost and impact of chunking or embedding. Preprocessing Cost: ~$30 for 68 sessions. Chunking/Embedding Cost: ~$0.75. Rationale: High-quality input data makes the choice of embedding model less critical.
- Architecture: Data: 130+ hours of transcripts. Preprocessing: Sonnet 4.6 & Kimike 2.5 for full-transcript analysis. Retrieval: Hybrid BM25 keyword search + vector embedding. Database: Local LensDB. UI: OpenWebUI with a custom Python filter function. LLM: Local GPT-OSS (via Olama) with a custom system prompt.
- Performance: GPT-OSS achieves ~80% of Opus's performance for ~1% of the cost, demonstrating the power of a strong retrieval system to compensate for a smaller model's limitations.
- Sharing: Patrick will package the processed LensDB and raw data for the community to download and use.
"AI Booth Studio" & Kiosk Engine
- Juan's Project: "AI Booth Studio" Concept: An on-site AI photo booth for events. Function: Guests select styles → photo transformed by inference engine → results delivered via QR code. Tech: Staging environment on an EC2 instance; authentication with Clerk.
- Morgan's Project: Kiosk Display Engine Concept: A robust, display-only kiosk engine for digital signage. Tech: Raspberry Pi running a full-screen Chrome instance. Function: Pulls content from a manifest, stores it locally, and displays it offline-first. Features: Content: AI-generated HTML slides from simple Markdown/JSON. Manifest: Controls slide timing and expiration dates. Resilience: Self-boots and recovers from failures.
- Synergy: Morgan's kiosk engine can be integrated with Juan's studio to create a dynamic public display of AI-transformed photos at events.
Other Project Updates
- Marc: Building investment agents using the Alpaca API, which have already generated $3,000 in simulated profit.
- Tom: Integrating a military association's systems (WordPress, Xero, Stripe) but facing a key challenge: no agreed-upon mechanism for creating membership numbers.
- Adam: Debugging Zoho CRM integration for a SaaS product.
- Patrick: Hardening Claude Code for an enterprise pilot by deploying org-level settings.json and a seeded cloud.md to constrain user behavior.
Next Steps
- Patrick: Package the "Community Brain" project (LensDB, raw data, instructions) for community sharing next week.
- Paul: Reinstall the Codex plugin in Claude Code and begin using Superpower to resolve a stalled project.
- Morgan: Share Raspberry Pi kiosk engine details (setup script, ISO) with Juan and Paul.
- Juan: Explore integrating the kiosk engine with the "AI Booth Studio" to add a public display feature.
Action Items
- Add pre-processing step to redact sensitive term in transcripts before Claude Code - WATCH (5 secs)
- Agree membership-number source w/ association secretary + Xero auditor; then add Xero custom field - WATCH (5 secs)
- Send LinkedIn invite to Morgan; then DM him re: kiosk hot-folder sync - WATCH (5 secs)
- Send LinkedIn invite to Morgan; then DM him re: conference kiosk screens - WATCH (5 secs)
- Reinstall Claude Code Codex plugin; then use Codex Rescue to debug nav map - WATCH (5 secs)
0
1 comment
AI Developer Accelerator — Coaching Call - May 5th
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
+11
2
+5
3
+3
4
+3
5
+1
Powered by