Fluorescent booth · technical screensVoice · video · rubric

Voice and video interviews · AI interviewer

Hear how they think.

Siftel is a core AI/ML product: a voice and video interviewer that follows your rubric. The interviewer works a problem with the candidate, flags cheating across keystrokes and camera, and returns a scored packet in minutes. A human still makes the offer.

  • Interviewer voice + video
  • Rubric cited scores
  • Integrity keystroke + camera
Claude 3.7 SonnetGPT-4oGemini 2.5Amazon NovaBedrockAnthropicOpenAI DirectBedrock Agents Claude 3.7 SonnetGPT-4oGemini 2.5Amazon NovaBedrockAnthropicOpenAI DirectBedrock Agents

AI features

Where the interviewer actually sits

Siftel is in the business of building a core AI/ML product for hiring. Named jobs: the live interviewer, rubric scoring, integrity flags, and the packet. This is an interviewer agent — a human still decides the offer.

AI feature · Voice and video interviewer

Works the problem with the candidate, live

For hiring teams who need a technical talk, not a multiple-choice form

Full-length technical talk with interruptions, follow-ups, and a shared editor. Ready at 2am as well as 2pm. Engineering, analytics, and ops roles.

Invite → live booth → POST /v1/screens

AI feature · Rubric scoring

Your skills. Cited evidence on each score.

For hiring managers who refuse a black-box grade

Paste the role, the stack, and three problems you actually use. Siftel cites quotes and editor moments for each skill. A human can still overturn the packet.

Rubric + session → GET /v1/packets/{id}

AI feature · Integrity flags

Keystroke, camera, and copilot patterns

For teams who need the booth honest

Flags cheating across keystrokes and camera. Unusual copilot patterns raise a flag. Integrity notes sit on the same packet as the scores.

Keystroke + video + audio → integrity channel

AI feature · Model-backed interviewer

Claude, GPT-4o, Gemini 2.5, Amazon Nova

For platform teams wiring the booth to models they already trust for accuracy

Planned model families: Claude 3.7 Sonnet, GPT-4o, Gemini 2.5, Amazon Nova. Access: Bedrock, Anthropic, OpenAI Direct. Agent frameworks: custom/homebrew and Amazon Bedrock Agents. Siftel is not fine-tuning those models.

Turn → selected model → Bedrock / Anthropic / OpenAI

HR / career services. Speech recognition and video. MVP stage — no invented customers, funding, or uptime claims. A human still makes the offer.

Method

How Siftel ships

Set the sieve. Invite the candidate. Read a packet a human can still overturn.

01

Set the sieve.

Paste the role, the stack, and three problems you actually use.

02

Invite the candidate.

They pick a slot. Siftel is ready at 2am as well as 2pm.

03

Read the packet.

Scores, quotes, and integrity notes. A human still makes the offer.

Stack

What you actually buy

Voice screen

Full-length technical talk. Interruptions, follow-ups, and a shared editor.

Integrity

Keystroke, video, and audio channels. Unusual copilot patterns raise a flag.

Rubric

You write the skills. Siftel cites evidence for each score.

ATS

Greenhouse, Lever, and Ashby sync. Packets land on the candidate card.

Start

A first screen this week.

Write founder@siftel.space. You get a test token, a sample screen, and a 30-minute walkthrough on the role you want to hire for.

See pricing