ShenSeanChen
On edge cases: the agent doesn't try to "fake" answers. If it doesn't have enough context from the website/knowledge base, it asks follow-up questions or flags the conversation as needing human review. We're also adding explicit escalation rules so certain intents immediately route to a human.
For handoff: the idea is that the agent handles the early conversation layer (answering questions, capturing requirements, qualifying). If the conversation becomes complex, the thread can be handed to a human with the full conversation history and extracted lead data.
On ICP: today we're mostly seeing traction with businesses that already receive inbound inquiries (often on WhatsApp) but don't have a structured way to capture or follow up on them. The agent sits in front of that chaos and structures it into leads.
Still early though — we're learning a lot from users right now.
Show HN: AutoManus MCP Server – create a sales rep agent from Claude in 1 min [video]
https://www.youtube.com/watch?v=4Mdzehq4ghg&list=PLE9hy4A7ZTmpGq7GHf5tgGFWh2277AeDR&index=37
Hi HN, I’m Sean, founder of AutoManus.We just released an MCP server + REST API that lets you create a sales representative agent for any business, directly from Claude Desktop or Cursor.The idea is simple: every AI product/agency/business needs a way to talk to customers, and you don’t want to build that layer from scratch every time. You want a sales layer, the sales interface to speak to the...
Show HN: Open Source MVP Launch Kit (NextJS, Stripe, Supabase), 563 Github Stars
https://twitter.com/ShenSeanChen/status/1895163913161109792
I'm open sourcing a template I built to launch a full-stack SaaS with Stripe Payment, NextJS, Supabase and Google OAuth, in ONE SHOT. It took me some time so I decided to share this to you all. The goal is to help everyone save time when launching your next SaaS, with EVERYTHING YOU NEED. You can build anything. You can learn anything.Links- Github Repo:...
The lagging is due to ElevenLabs API calls and slower internet at home using hotspot. When using self-hosted TTS models the lagging is negligible.
Thanks again for your feedback!
The problem: 1. Cooks struggle with messy hands while trying to follow recipes. 2. Existing voice assistants can't navigate complex recipes or set context-aware timers. 3. Phones and printed recipes become unusable with greasy fingers.
Technical solution: 1. Voice-controlled recipe navigation with natural language understanding. 2. Dynamic timer setting based on recipe steps. 3. Cross-device sync using WebSocket connections. 4.Attempted with several speech-to-text techs and a custom NLU layer (Python/Next.js).
Key challenges: 1. Handling ambiguous voice commands in a noisy kitchen environment 2. Synchronizing updates when having multiple timers 3. Normalizing recipes from various formats (PDFs, websites, handwritten scans)
Current stack: 1. Supabase for real-time database 2. Custom voice pipeline (very low latency, check out this live demo video: https://youtu.be/XKxoHOIxyjA)
I'd appreciate feedback on: 1. Preferred recipe import formats 2. Commonly used voice command patterns while cooking 3. Additional timer management / task-performing features you'd find useful.
There's a 7-day free trial available. I'm around to answer questions and discuss the technical aspects of the project.I'm around to answer questions and discuss the technical aspects of the project here: 1. https://x.com/ShenSeanChen 2. https://discord.gg/TKKPzZheua 3. https://linkedin.com/in/shen-sean-chen