Drift Quotient

AI-powered organisational analysis platform

Problem

Organisations routinely make bold claims about their culture, values, and performance that don’t match reality. Investors, board members, job seekers, and partners need rapid, evidence-based assessments of organisational authenticity – but traditional due diligence takes days or weeks of manual research, relies on subjective interpretation, and often misses critical evidence from employee reviews, regulatory filings, and independent assessments.

Website dashboard screenshot, a grid of white cards with company names written in, each card links to an analysis report page.

Solution

EthosSignal approached me with a unique organisational analysis framework documented in Notion. They needed me to transform their “Drift Quotient” scoring methodology into a fully functional production platform within 6 weeks. The challenge was to build an AI-powered system that could automate multi-source evidence gathering, detect contradiction patterns across five organisational dimensions, and deliver actionable insights – all while maintaining evidence transparency and cost efficiency.

Website screenshot showing a report summary and status page, a table with 6 rows shows results and status of individual analyses

Approach

I architected and built the complete platform from scratch using Next.js 15 with React Server Components, OpenAI GPT-5-mini with web search capabilities, Supabase for authentication and PostgreSQL storage with Row Level Security, and PostHog for comprehensive LLM observability tracking costs, tokens, and latency per API call.

The core technical challenges I solved included: designing a parallel execution system to run 5 AI analyses simultaneously reducing total analysis time from 12–15 minutes to 4–6 minutes; implementing structured AI outputs using Zod schemas for type-safe parsing and validation; engineering specialised prompt templates with signal detection frameworks for absolutist language, tone mismatches, and direct contradictions; building a user-suggested URLs feature allowing users to provide specific evidence sources; and creating real-time status tracking with elapsed timers and state management to prevent duplicate runs.

Website screenshot of a organisational drift report, a bar visualises result from high, medium or low drift, and boxes have sentances and paragraphs written inside with insight into why the score was given

Results

I delivered the production-ready MVP on schedule with complete authentication flows, organisation management, a distraction-free analysis dashboard with real-time status updates, shareable public report pages with colour-coded drift scores and evidence citations, and full LLM observability. Each complete analysis costs approximately $0.05-0.15 thanks to GPT-5-mini prompt optimisation & reasoning, and efficient parallel execution – launching an AI-powered analysis platform that delivered immediate value while maintaining cost efficiency at scale.

🕐 50-100x faster analysis than manual research (6 minutes vs. 5-10 hours)

🔎 3-5x more evidence coverage (30-50+ sources vs. 10-15 manual finds)

☑️ Standardised scoring methodology that’s repeatable and defendable

View the project