Project ATS

Hiring was running on spreadsheets and a thread of forwarded emails.

The talent team was hiring across four functions with the tools you'd expect from a five-person startup, not a growing org, Google Sheets for the pipeline, Calendly for scheduling, separate docs for every interview, and a Slack channel where scorecards went to die. Two recruiters could call the same candidate in the same week, neither knowing the other was already mid-loop.

Worse, every hiring manager wrote their own interview questions from scratch. A backend role and a frontend role were evaluated against entirely different bars, so 'who's the strongest candidate' became unanswerable the moment more than two people had opinions.

The brief was a single internal ATS that enforced structure without slowing the team down, define the role once, generate the question set, score everyone the same way, and surface the offer-ready candidates without a 14-tab spreadsheet.

Define the role in four steps, let the model write the questions.

Every job starts with the wizard. Step one, Annual Objectives, captures the 3-5 outcomes the role exists to deliver. Step two ties those to Quarterly Goals so the rubric is grounded in the work, not the title. Step three is where the system earns its keep: recruiters answer four input prompts per competency (Skills, Experience, Problem Solving, Culture Fit), and the model generates the interview questions in seconds, regenerate or approve, no blank-page paralysis.

Step four locks in compensation bands and ideal-background notes. By the end, the role has a structured brief, a question library, and an evaluation rubric, generated once, reused across every candidate in the loop.

A pipeline that gates each round, and a scorecard the whole loop shares.

Every job opens to a candidate pipeline organised by round, Ideal Career Profile, Evaluation Criteria, Work Simulation, References. A round can't be marked complete until its scorecard is filled, so candidates don't quietly skip stages and surface at offer-time with half a rubric.

Inside Evaluation Criteria, the four competency tabs (Skills · Experience · Problem Solving · Culture Fit) carry the AI questions generated upstream. Interviewers capture the candidate answer summary, their own thoughts, and a 0–10 score per question. The Total Score rolls up at the bottom, same rubric for every candidate on the role, no more apples-to-oranges debates.

Compare candidates on the same rubric, and schedule without leaving the tool.

Because every candidate on a role is scored against the same generated question set, ranking is a side-effect of the data, not an end-of-loop debate. The pipeline view sorts by Total Score, surfaces the offer-ready set, and flags rounds where two interviewers diverge by more than two points so the team can resync before extending an offer.

Interview scheduling is wired in: invite the candidate, pick the interviewer, and the platform sends the calendar invite plus reminder emails through SendGrid, no Calendly handoff, no double-booking, no missed loops.

One dashboard for hiring KPIs, open roles, and team load.

The admin view is the talent lead's morning check-in. Open roles, applications per role, time-in-stage averages, and interviewer load across the team, all surfaced in a single screen so bottlenecks (the role stuck in 'Work Simulation' for three weeks, the interviewer with eight loops on their plate) become visible before they become problems.

Job listings, candidate exports, and round-level analytics are exposed via the same dashboard, so the team runs hiring on data instead of vibes.