ACIA product · Live by invitation

InferaNoteInferaNote

Record a meeting in English or Mandarin, or import a recording that mixes both, and InferaNote writes the note, puts the action items on your agenda, and links the people, organisations and decisions into a knowledge graph you can ask questions of.


Person Organisation Decision Action Date Document⟲ provenance: the moment in the recordingIllustrative example: names are not real meeting data. The centre shows the source meeting; in InferaNote each relation records that meeting as its provenance rather than linking to it as a node.
Knowledge Graph

Every Meeting Joins One Graph

The people, organisations, projects, events and dates a meeting mentions become nodes in one graph. Ask a question and the answer is drawn across every meeting.

  • Relations, not tagsattends, discussed, responsible_for, scheduled_on, cites: a controlled vocabulary, so the graph stays queryable as it grows.
  • Provenance on every edgeEach relation records the meeting, document or note it came from and, for a recording, the moment it was said where that can be found, so a claim in the graph can be traced back to its source.
  • Curated cross-language aliasesEnglish and Chinese names for the organisations on a curated alias list fold into one node, so NVIDIA is one node whichever script a meeting used.
  • A page per entityEvery node has its own page listing what was decided, promised and cited about it, meeting by meeting.

How It Works

Three Steps, One Corpus

  1. 1. Record or importRecord a meeting or import a recording, and upload documents or photos from your phone.
  2. 2. Note and agendaA structured note is written with a summary, decisions, action items with owners and key dates, and the action items land on your agenda.
  3. 3. Graph and assistantThe people, organisations and dates join the knowledge graph. Ask the assistant by text or voice: answers are grounded in your own notes and documents and cite the moment in the recording they came from.

The AI Layer

Built on AWS, End to End

InferaNote is a whole system rather than a single model call. Its assistant runs on a choice of Bedrock models and gives spoken answers through Amazon Polly.

  • English and Mandarin in one imported recordingLanguage identification between English and Mandarin on Amazon Transcribe, so an imported meeting that moves between the two languages is transcribed in both rather than forced into one.
  • Notes and extraction on BedrockAnthropic Claude on Amazon Bedrock writes the note and extracts the people, organisations, projects and dates it mentions.
  • Retrieval that stays honestAmazon Titan embeddings with an exact sweep over the curated corpus, and an approximate S3 Vectors index only for bulk archives — every approximate hit is labelled as such at the citation.
  • Learning loopsSpeaker labelling with playable samples, a custom vocabulary that improves transcription of names and terms, and a style profile that only becomes memory once the owner confirms it.
Read the full product page on acia.org.au ↗

Target architecture · Gateway in development

InferaNote and the InferaStack Product Line

WorkloadInferaNoteNotes · assistant · graph, in an AWS account in ACIA's AWS organisation
→
Workload interfaceInferaStack GatewayWhere OpenAI-compatible access, per-key budgets and a metadata-only audit record would sit in a customer deployment; in development, not commercially released
→
AI servicesAWSClaude on Amazon Bedrock for notes and extraction · Titan embeddings · Transcribe for English and Mandarin speech · Polly for spoken answers
→
Compute placementInferaStack AI GridWhere sovereign capacity would serve this workload later; proof of concept today

The flow shows where the InferaStack product line would sit in a customer deployment: the InferaStack Gateway in front of the models, which is in development and not commercially released, and InferaStack AI Grid capacity underneath, which is at proof of concept. It is a target architecture, not a description of how InferaNote runs now.

What is live

InferaNote for invited ACIA members, calling AWS AI services directly.

What comes later

Placement on InferaStack AI Grid capacity, only if the grid passes its proof-of-concept and site gates and a commercial decision follows.


Privacy and Ownership

A second brain is only useful if it is yours. These are the rules it runs under.

  • AWS account and RegionInferaNote is deployed in an AWS account in ACIA's AWS organisation, and its data is stored in the AWS Region the owner chooses. The AI layer uses AWS AI services from that account; Bedrock models are called through cross-Region inference, so inference can run in other Regions, and each deployment documents where every model runs.
  • Private by defaultInvited members only — ACIA members and designated users. No open registration, and no public content.
  • Export any timeNotes, meeting summaries, entity pages and document text export as plain Markdown with wiki-links, so your notes are not locked in.
  • Notion is one-wayA note can be published to Notion. Nothing is read back from it.

See InferaNote

ACIA owns and operates InferaNote, holds its data in an AWS account in ACIA's AWS organisation and decides who is invited. InferaNote is run by ACIA and is available by invitation to ACIA members and designated users. The product page on acia.org.au has the full feature list and current architecture. To bring InferaNote to your team, ask ACIA.