← Back to all projects
Project 01

Deep AI Training at Scale

Full infrastructure for structured voice sessions between human experts and AI agents — multi-platform, fully processed, team-briefed in minutes.

200+
participants
115+
call time in hours
45
sessions
600+
structured questions
6k+
training data pairs
<5min
session to Slack brief
100%
auto-processed

Five layers. Fully automated.

Voice Layer
Retell AI
VAPI
Bland AI
Orchestration
n8n Cloud
Analysis
Assembly AI
OpenRouter
Claude
Storage
Airtable
Supabase
Frontend
Vercel
Dashboard
How It Works

Capture. Analyze. Brief. Repeat.

Segment The Work
Voice Platform AI voice agent runs Socratic sessions with participants — goal-directed, no human facilitation.
n8n Webhook fires on call end. Ingests participant ID, recording URL, and metadata into the pipeline.
Assembly AI Full transcription with speaker diarization, timestamps, and sentiment markers.
OpenRouter / Claude Scores the session, extracts key moments, and packages turns into SFT, DPO, and CONVICTION SHIFT formats.
Slack + Airtable Structured brief posted to Slack. Full record written to Airtable within minutes of call end.
Frontend Custom dashboard to review sessions, track participants, monitor scores, and manage the study in real time.
Hear It

Real sessions. Real participants.

Actual recordings from LifeTrade training sessions — the AI voice agent running Socratic interviews, processed automatically after every call.

Training Interview — Sample 01
Deep AI Interview · Growth Domain
7:48
0:00
7:48
Training Interview — Sample 02
Socratic Interview · Knowledge Extraction
10:43
0:00
10:43
Training Interview — Sample 03
Extended Session · Multi-Turn
24:36
0:00
24:36
Training Methodologies

One session. Multiple formats.

Every session is automatically packaged into the formats AI labs need most — structured at the point of processing, not after.

SFT
Instruction-response pairs from Q&A turns
DPO
Ranked preference comparisons from session turns
KTO
Good/bad labeled outputs with human ratings
CONVICTION SHIFT
Before/after belief revision with numerical ratings
GCAI
Goal-directed reasoning chains from session flow
SPIN
Self-play traces from multi-turn sessions
PROCESS
Step-by-step reasoning annotation by reviewers
Stack

What it runs on.

Voice AIRetell AI
Voice AIVAPI
Voice AIBland AI
Orchestrationn8n Cloud
Voice AnalysisAssembly AI
AI AnalysisOpenRouter / Claude
Session StorageAirtable
NotificationsSlack
Team DashboardVercel
AI ModelMistral
GPU InfrastructureRunPod
Serverless ComputeModal