ClearHire-AI

ClearHire turns a 36-day hiring process into a 2-second AI decision — powered by UiPath Maestro BPMN, Groq AI, and WhatsApp

Comment

Inspiration

As a student applying to internships and roles myself, I know the feeling of submitting an application and hearing absolutely nothing back — no rejection, no confirmation, just silence. Turns out this isn't just my experience: 52% of candidates never hear back after applying, and the average time-to-hire sits at 36 days. Meanwhile, recruiters spend roughly 80% of their time manually screening candidates they'll ultimately reject anyway.

We wanted to fix both sides of this broken loop — give recruiters their time back, and give candidates an actual answer, fast, on a channel they actually check. For most of the world outside North America and Europe, that channel is WhatsApp, not email. So we built ClearHire around that insight.

What it does

ClearHire is an AI-powered recruitment orchestration system that takes a candidate from CV submission to hiring decision in under 2 seconds.

A candidate applies through our portal and uploads their CV. Behind the scenes, UiPath Maestro BPMN kicks off an orchestrated workflow: it calls our AI scoring engine (Groq running Llama 3.1 70B) to evaluate the CV against the job's requirements, hits an Exclusive Gateway to make a routing decision based on the score, then updates the candidate's record and fires off a WhatsApp notification — shortlisted or rejected, either way the candidate knows immediately.

For borderline scores, the workflow is designed to route to a human reviewer via UiPath Action Center rather than letting the AI decide alone — keeping a person in control of the calls that actually require judgment.

HR gets a real-time dashboard showing every candidate's AI-generated score, extracted strengths and gaps, and full status history, with the ability to override any automated decision at any time.

How we built it

  • Orchestration layer: UiPath Maestro BPMN models the entire recruitment pipeline as a business process — Start Event → AI Scoring Service Task → Exclusive Gateway → Shortlist/Reject paths → status update and notification tasks → End Event. Three UiPath API Workflows connect Maestro to our backend over HTTP.
  • AI layer: A Node.js/Express backend on Google Cloud Run extracts text from uploaded CVs, builds a structured prompt, and sends it to Groq's Llama 3.1 70B model, which returns a JSON score (0–100) plus strengths, gaps, and a recommendation.
  • Data layer: MongoDB Atlas stores Job and Candidate documents, including a full audit log of every status change the BPMN triggers.
  • Notification layer: Meta's WhatsApp Cloud API sends candidates real-time updates using purpose-written message templates for every stage of their journey.
  • Frontend: A React + Vite dashboard gives HR full visibility and manual override control over every AI decision.
  • AI-assisted development: Architecture design, BPMN flow logic, Groq prompt engineering, and the MongoDB schema were developed with Claude AI (Anthropic) as a coding agent throughout the build — fully documented with prompt logs in our /ai-development-log folder in the GitHub repo.

Challenges we ran into

Getting UiPath Maestro's HTTP Request activities to reliably pass dynamic workflow variables into request bodies took real trial and error — we ended up standardizing on a clean, explicit variable-mapping pattern between API Workflow inputs/outputs and the BPMN process variables. We also hit API quota limits during testing and had to build a resilient fallback layer so a rate-limited AI call never breaks the orchestration mid-flow. Designing a scoring prompt that reliably rejected non-CV documents (so the system can't be gamed by uploading random files) also took several rounds of prompt refinement.

Accomplishments that we're proud of

We built a working, end-to-end orchestrated system — not a slide deck — where a real candidate action on WhatsApp triggers a real UiPath BPMN process that runs in under 2 seconds and produces a real, auditable hiring decision. We're also proud of designing a human-in-the-loop checkpoint into the flow rather than letting AI fully replace recruiter judgment.

What we learned

UiPath Maestro is genuinely powerful as a governance layer over AI decisions — it's not just an automation tool, it's an audit trail and a control point. We also learned a lot about prompt engineering for structured, reliable JSON outputs that downstream systems can trust without manual validation.

What's next for ClearHire-AI

Adding Document Understanding for parsing scanned/image-based CVs, building out the full Action Center escalation path for borderline candidates with SLA timers, and expanding WhatsApp into a two-way channel where candidates can ask questions about their application status directly.

Built With

Share this project:

Updates