DS
Diwesh Saxena
50% ↓ time-to-hire

AI-powered ATS: 50% reduction in time-to-hire

Chief Technology Officer•Noble House•2019–Now

Built a modular ATS with resume parsing, ranking, and analytics. Added eval harness and hiring funnels to halve time-to-hire.

Technology Stack

Python.NET CoreReactPostgreSQLAzure/AWS

Key Outcomes

  • •50% reduction in time-to-hire across the Noble House consultant pipeline
  • •Improved candidate NPS through transparent, faster hiring stages
  • •Real-time funnel analytics enabling data-driven recruiter decisions
Outcome: 50% ↓ time-to-hire with an AI-augmented ATS.

Context

Noble House needed hiring to move at the speed of business. Legacy tools created latency and blind spots across sourcing, screening, and scheduling. We set a singular KPI: cut time-to-hire by half while increasing candidate satisfaction.

Problem

  • Recruiters were juggling multiple systems; data-entry and follow-ups slowed offers.
  • Screening lacked consistent, fair scoring; interviews clashed; funnel drop-off was invisible.

Constraints

  • Integrate with existing job boards/HRIS.
  • Preserve auditability and reduce bias; support remote-first teams.

My role & team

I served as CTO and architect, leading cross-functional squads (product, data, platform, QA). I defined the KPI model, architecture, and an evaluation harness for ranking quality.

Approach

  1. Discovery & KPI model — mapped current funnel and baseline SLAs.
  2. Data pipeline — resume ingestion and canonical candidate profile.
  3. Ranking & rules — skill signals, recency, constraints; human-oversight loops.
  4. Scheduling & comms — automated calendars, SMS/email nudges.
  5. Funnel analytics — time-in-stage, drop-off, and alerting.
  6. Evals & guardrails — weekly tests for precision/cost/latency; bias checks.

Architecture (at a glance)

Ingestion (parsers/webhooks) → Profile Store (PostgreSQL) → Ranker API (Python) → Orchestrator (.NET) → UI (React) → BI (funnel dashboards) → Notification service (email/SMS).

What we built

  • Resume parsing + profile unification with manual overrides.
  • Ranking service combining rules + learned signals; HITL review queues.
  • One-click scheduling with calendar sync and candidate preferences.
  • Funnel dashboards with SLA alerts; offer workflow with approvals.

Results

  • 50% ↓ time-to-hire and faster recruiter response times.
  • Higher candidate satisfaction; fewer drop-offs at screening/interview.
  • Reliable reporting for leadership and clients.

Lessons

  • Data hygiene beats model cleverness.
  • Guardrails + evaluation harness keep models honest as volume grows.

Next

  • Integrate structured interview rubrics; expand fairness audits; add sourcing marketplace.

CTA

Want a deep dive into the evaluation harness or ranking rules? Book a 30-min call.