We originally built a marketplace connecting startup advisors with founders. After ~6 months, we had hundreds of recorded discovery calls and transcripts.
We noticed something interesting: The first 10 minutes of most advisory calls were almost identical. • What do you do? • Who is this for? • Pricing? • What information do you need from me? • Am I a good fit?
So we pivoted.
Today, MyClone lets service professionals (consultants, CPAs, etc.) deploy an AI voice agent that: • Answers FAQs in their voice • Runs structured intake conversations • Qualifies leads • Routes good prospects to a human
Technical details
Under the hood: • Multi-source ingestion (websites, docs, PDFs, media transcripts) • Structured knowledge graph + chunked semantic retrieval • Custom RAG pipeline (not using off-the-shelf frameworks) • Prompt templates generated dynamically based on objective (FAQ vs intake vs qualification) • Guardrails to prevent hallucination beyond uploaded corpus
We learned quickly that generic “chat with your docs” doesn’t work for client-facing scenarios. The hard part is: • Controlling tone • Avoiding overconfidence • Structuring conversations instead of answering open-ended questions • Handling partial / messy user inputs • Maintaining low latency in voice mode
Still early, but we’re seeing adoption from solo and small firms who want a 24/7 “AI associate” for first-touch conversations.
We do share our technical learnings in our blog - https://www.myclone.is/blog Happy to answer technical questions, RAG architecture details, failure modes, or things that broke during the pivot.
— Viggy