Rohit: Fixed 3 bugs — file attachment without text now works, journal test globe fixed, session display issue resolved.
Rahul: Site settings UI complete with labels, icons, age validation, dynamic quotations. Onboarding summary updated. Backend still needed for signup/subscription.
Mukesh: Onboarding APIs updated for admin + user side. Working with Sumit on AWS IMV credential vault config. Expected complete by tomorrow EOD.
Phase 1 complete using AIORTC library in Python. Demo working but not production-grade (noise handling, edge cases). Jason asked for LiveKit comparison — LiveKit is turnkey/production-ready while AIORTC is more customizable. Sumit estimates production-ready backend by Friday, frontend integration by Tuesday next week. Voice agent needs to be embeddable as HTML widget for use in publisher and experiences.
Jason demoed the MasteryMade publisher dashboard: page management, side-by-side code/preview viewer, image upload with auto-resize, YouTube embed, drag-and-drop page creation, version history with rollback, and multi-domain support. Goal: publish anything to any domain instantly with tagging and folder organization. Eventually auto-generate menus from page collections.
Demonstrated turning articles into 3-part lead magnets: (1) editorial overview, (2) step-by-step playbook with AI prompts and "good enough" criteria, (3) locked implementation workbook requiring email capture. Template is reusable for any article. Sumit suggested adding share buttons to sticky header. Concept: "dynamic lead magnets" built from any expert content.
Phase 1 complete. Creates PRD → generates plan → opens containerized sessions to code each part → tests → fixes in new containers → loops until done. Key innovation: runs in separate containers to avoid context window limits. Adding budget estimation: calculates token cost for plan/build/test/reimplement phases with 30-50% buffer. Users set spend limits before autonomous execution in "risk mode" (no permission prompts).
Four-layer concept: (1) Personal briefing — monitor YouTube/Twitter/email subscriptions, weekly market summary. (2) Content pipeline — monitor keywords for client niches, create writing briefs, AI agents write articles, auto-publish. (3) Self-monitoring — ingest own published content into vector DB for consistent voice/references. (4) Client auto-update — when Bridger publishes new content, auto-update his knowledge base without manual intervention.
Sumit exploring Codex (ChatGPT desktop) as alternative to Claude/Cursor. Jason recommended Gemini CLI for its generous free tier (~500-1000 queries/day). Strategy: use Claude Opus for planning/PRD (best engineer), cheaper models for coding. Gemini strong on frontend, Claude on backend. Always have 2+ windows open — CLI for coding, chat as assistant. Switch between paid services to conserve tokens.