Nova Hiring — Jorge Vázquez Moreno

01 — Context

Manual screening was the bottleneck killing recruiter productivity

Nova Hiring targets companies with high-volume blue-collar hiring needs — logistics, hospitality, retail. The founding team had strong technical capability but needed product thinking to translate a promising AI concept into something recruiters would actually use.

The core insight was simple: recruiters were spending 60 to 70 percent of their time on initial screening calls that followed predictable, repetitive patterns. That was the problem worth solving.

02 — Problem

The AI existed but the product did not

The team had a working prototype that could parse CVs and generate candidate summaries. But it was not a product — it had no defined ICP, no clear workflow integration, no way to measure quality of shortlists, and no feedback loop to improve the model outputs over time.

My job was to turn a technical demo into a product that solved a real recruiter problem end to end.

◎

The missing piece

Speed alone was not the product. Recruiters needed to trust the shortlist. Without a quality signal, even a fast pipeline would be ignored — recruiters would re-screen manually to verify what the AI had done.

◈

The feedback loop gap

The prototype had no mechanism to learn from recruiter decisions. Every shortlist existed in isolation. Without closing the loop between AI output and human judgment, the model had no path to improvement.

03 — Key decisions

What I chose to build, and why

01

Defined the ICP tightly I defined the ICP as HR teams of 3 to 15 people in companies hiring 50 or more blue-collar workers per month. This narrowed the problem enough to design a focused workflow rather than a generic AI tool.

02

Async screening via WhatsApp conversational interviews The pipeline was designed around structured async screening: candidates received a WhatsApp-based conversational interview driven by a prompt architecture I designed. This removed the need for a human on the first call entirely.

03

Shortlist quality score to close the feedback loop I introduced a shortlist quality score — a recruiter-facing metric that rated each AI-generated shortlist based on how many candidates advanced to the next stage. This created the feedback loop the model needed to improve.

04

Zero technical setup for recruiters Onboarding was designed for zero technical setup. Recruiters could define a role, set screening criteria in plain language, and launch a campaign in under ten minutes.

04 — Results

Speed and quality, not just speed

75%

Reduction in time-to-shortlist compared to manual screening baseline

10min

Time for a recruiter to launch a new screening campaign from scratch

0

Human calls required in the first screening stage

✓ Quality

Recruiters reported shortlist quality equal to or better than manual screening in early user interviews

05 — Reflections

What I would do differently

01

Instrument shortlist quality from day one We should have instrumented shortlist quality from day one, not added it later. The first two months of data were lost because we had no way to close the feedback loop on model outputs. That data gap cost us iteration cycles.

02

Scope a parallel web-based flow earlier The WhatsApp channel was the right call for blue-collar candidates, but it created compliance questions in some markets that we had not anticipated. A parallel web-based flow should have been scoped from the start as a fallback.

An LLM pipeline that cut hiring time by 75%