Silta is a SaaS platform for investment professionals evaluating renewable energy projects. Their analysts work through data rooms of 100–200 documents per project, but during live calls with project stakeholders, there's no way to search that material in real time. Our challenge was to put the data room inside the meeting.
When an investment analyst is on a live call with a project developer, a technical question comes up, something buried in a 200-document data room, and the analyst has seconds, not minutes. Silta came to us to validate whether that was technically possible: a meeting companion that joins a Google Meet, detects questions as they happen, and delivers answers with sources before the moment passes.
Real-Time Meeting Bot
Pedro Assistant joins a Google Meet as a participant via a self-hosted attendee bot running headless Chrome. It streams live audio to Amazon Transcribe, which finalises transcript segments after each natural pause. Segments are passed to Amazon Bedrock, which classifies whether a question was asked.
Sub-10-Second Answer Delivery
When a question is detected, the system queries the Silta API — backed by Pinecone vector search — and surfaces a source-cited answer in the React UI in under 10 seconds.
Session Orchestration
The backend runs on EC2 with Docker Compose. Redis stores transcript chunks, detected questions, and session state. The entire AWS environment is now managed by Terraform.
Pedro Assistant proved the thesis: a real-time AI meeting companion can listen to a live investor call, detect questions, and retrieve source-cited answers from a 100–200 document data room in under 10 seconds. The validated architecture gives Silta a foundation to evaluate integrating Pedro into the product.
Tell us more about yourself and what you're got in mind.
Prefer email? hello@apexlab.io
Or even a quick call?
We don’t care about fancy CVs or long interview processes. Provide some basic information, let’s have a quick chat, and then show us what you love doing the most: write some code and create a product.
