Many conversations. One record.

Have the conversation. Kayda keeps the record. Built first for employee relations investigations.

One interview ends up in eight places

An investigation interview can decide a career, and its finding can end up before the Fair Work Commission. Yet the record of what was said is spread across tools that were never built to hold it.

Today

The interview

Zoom or Teams, Recording expires
Transcription app, Speakers mislabelled
Word, Typed from memory
Email, Buried in a thread
Chat, A screenshot, unverified
Ticketing system, Re-typed by hand
Shared drive, Hard to find later
HR system, No record of what was said

One conversation. Eight places. No single record.

In Kayda

It stays one record

Sources

booking-9-may.pdf, SHA-256 kept
policy-extract.pdf, SHA-256 kept
Each line cited, to its file and passage

Conversation

Notice agreed, on screen, wording kept
Recorded, inside the case
Transcribed, as it runs

Suggestion

Bears on allegation 1, quoting the line
Still to establish, how he knows Dave
Suggested by AI, not evidence

Record

Read back, and confirmed
Sealed, by a named person
Record of investigation, built from the case

ER-2026-014, one record

When a finding is challenged

The employer answers with a manager's recollection and a folder of files.

With Kayda, it answers with the line as it was said, who agreed to the recording, and the sealed transcript.

From the files you have to a sealed transcript, in one case

Kayda's first product is for employee relations investigations. Meetings and research run on the same record engine, in development.

Drawn from Kayda's screens.

The file

Start from the files you have

Say what happened. Add your files and a sentence or two. Kayda drafts the parties, the allegations, a dated chronology and an evidence register, and every drafted party, allegation and chronology line quotes the passage it rests on.

Check every line. Nothing is filed until you have confirmed, edited or removed each line. Every file is kept with its SHA-256 hash.

What you say

“Callum Weir put a private bucks party at Harbourside Social on his corporate card as client entertainment.”

Your files

booking-9-may.pdf
policy-extract.pdf

What Kayda drafts

Suggested by AI, not evidence

Party
Callum Weir, the respondent
booking-9-may.pdf, page 1: “booked as Callum W”
Chronology
15 May 2026, Dave's bucks in Karaoke Room 2
booking-9-may.pdf, page 1: “Karaoke Room 2 for 15 May”
Allegation 1
Bucks party claimed as a client dinner
booking-9-may.pdf, page 1: “Dave's bucks, 12 guests”
Policy
May bear on Travel and Entertainment Policy 7.6, private celebrations
policy-extract.pdf, page 1: “Private celebrations, including birthdays, weddings and the events leading up to them, are not client entertainment”

The interview

Listen. Don't type.

Agree the notice first. Everyone who joins, a support person and an interpreter too, agrees to the recording notice on screen before they are let in, and each agreement is kept with the notice's wording, version and time.

Hold it inside the case. The interview is recorded and transcribed as it runs, so you can stay with the person, not the keyboard.

Or bring it in. Held it in a room or on Teams: bring in the recording, the transcript or your notes, and it joins the case.

Notice agreed Callum Weir, Recording notice, version 3, 21 May 2026, 10:02

Transcript 00:46 Callum Weir Dave is a contact.

The suggestion

See what is still to establish

While it runs. Kayda suggests which allegation a line bears on and what is still to establish, quoting the line. You ask every question in your own words.

After it. The summary for each allegation cites the lines it rests on, for the interviewer to approve or correct.

Suggested by AI, not evidence

00:46 Dave is a contact.

Establish how Mr Weir knows Dave, since the booking on file names the evening Dave's bucks.

Bears on A1

The investigator asked in their own words at 01:05.

Suggested by AI, not a finding

Summary for A1

Mr Weir described Dave as a contact. 00:46

The record

Sealed by a named person

Review and read back. You correct and redact the transcript; the interviewee reads it back and confirms or queries it.

Seal at Finalise. A named person seals each transcript, so a later change can be found. The record of investigation is built from the case and exports as Word or PDF.

ER-2026-014, interview with Callum Weir

00:46Dave is a contact.

Sealed at Finalise by Priya Raman, 23 May 2026

Sealed

The findings are yours to make. Kayda keeps what they rest on.

See it on a practice case

What the case shows when a finding is challenged

Who agreed, and to what

Each person's agreement to the recording notice, with its exact wording, version and time.

What was said, and by whom

Every line with its speaker and time, corrected by the investigator and read back by the interviewee.

Suggested, never decided

Each AI suggestion is flagged as one and kept apart from what people decided. People make every finding.

Who decided

Each finding and each outcome is recorded under the person who made it, with the date.

Every change, logged

Each change a signed-in person makes to the case is written to a log that can only be added to, with what changed and when.

Sealed, and checkable

A named person seals each transcript at Finalise, so a later change can be found. Each Word or PDF export is logged with who made it and its SHA-256, and Check a copy says whether a file is an unchanged export.

Where your cases live

Stored in Sydney

Case records, recordings and files are stored in Sydney, with Supabase on AWS, and Kayda runs in Vercel's Sydney region.

Walled off

Each organisation's records and files are kept apart from every other's in the database.

Transcribed in Australia

Speech is transcribed on Deepgram's Australian service.

Team

Daanish Duvvi

Co-founder, employee relations

10 years in HR and employee relations, now a Senior Employee Relations Specialist at Qualtrics, after Fiskars Group, Accenture and Tetra Tech.

Cankut Eker

Co-founder, research and engineering

PhD in physics, University of Wollongong. Built Kayda's platform, after an in-house agentic framework that brought an IT team's whole software stack into one place.

See Kayda on a practice case

Tell us a little and a founder will reply by email to set up a practice case, with invented people and events, so no real case details are needed. Design partners, investors, and anyone reviewing Kayda for IT, privacy or legal are welcome too.