# Smarequ — Recruitment Lifecycle Platform

An autonomous AI recruiter on Google ADK. Processes resumes, scores candidate fit, and orchestrates interview pipelines without losing recruiter judgement.

## Key facts

- **Domain:** HR & TALENT
- **Status:** In production with one client
- **Use when:** Enterprise hiring teams want to move from manual screening to LLM-orchestrated funnel without losing recruiter judgement.
- **Reference engagement:** Available for new engagements

## The problem

Enterprise hiring is broken in a familiar way. Recruiters manually triage hundreds of resumes per role. The 80 per cent rejected on first pass take more time than the 20 per cent that proceed. Quality candidates fall through. Time-to-hire balloons. The recruiters doing this all day are the most expensive humans in the funnel.

## The approach

We built an autonomous recruiter on Google ADK with progressive-disclosure Skills inspired by the Anthropic open standard. Three tools fire in sequence per candidate: transcribe the PDF, generate a candidate profile against the job spec, and decide on application status. The agent is designed to compress per-candidate processing from hours of recruiter time to minutes.

## The architecture

Three Cloud Run services on GCP. The Core API (FastAPI) handles the REST surface for the React dashboard. The Agent Service runs the ADK agent under the GoogleProvider with Gemini 3 Flash. The Interview Worker subscribes to Pub/Sub and processes interview analysis asynchronously. State lives in Firestore with a denormalised summary map for O(1) list reads. Resumes upload to GCS with an Eventarc trigger that fires the agent on object creation. Skills auto-discover from the agent's skills directory, with role-specific scoring heuristics for Technical, UX Design, Sales, and Operations roles. Auto-shortlist and auto-reject thresholds live on the Firestore job document, configured by the recruiter via a Blueprint editor; the agent reads them at runtime, with no hardcoded thresholds anywhere in the agent or skill layer. The dual-workflow pattern is the non-obvious choice: fresh uploads run a direct three-tool pipeline with zero LLM orchestration overhead for the happy path; partial or failed states route to the ADK agent with full LLM orchestration for graceful recovery.

## The outcome

A recruitment funnel built to compress hours of triage into minutes. Recruiter time shifts from triage to high-value conversations. An audit trail per decision. Role-specific scoring rubrics that the recruiter can tune without touching code.

## ATI shape

- Predictive intelligence — Scoring models for candidate fit, dimension-weighted by role
- Agentic execution — Google ADK orchestrating three tools in sequence, with LLM-driven recovery for failure states
- Secure data — Firestore + GCS in the customer's GCP project, with no data leaving their perimeter
- Actionable outcomes — Automatic shortlist, reject, or review decisions written back to the application document

## Tech stack

- Python
- Google ADK
- Gemini 3 Flash
- Gemini 3 Pro
- FastAPI
- React 18
- TypeScript
- Firestore
- GCS
- Eventarc
- Cloud Run
- Pub/Sub
- Anthropic Skills

## Impact

- Minutes — Per-candidate processing, down from hours of recruiter triage
- Zero — Hardcoded thresholds (all recruiter-configurable)
- Dual-path — Direct pipeline + ADK recovery
- Skills — Role-specific evaluation rubrics

## Related

- [All accelerators](https://levent.ai/accelerators/)
- [Next: Maestro](https://levent.ai/accelerators/maestro/)

Metrics on this page are estimated and expected improvements describing the design intent of the accelerator. Real-client delivered metrics stay in private decks; see https://levent.ai/ai-content-policy/ for the abstraction policy.

---

**Canonical URL:** https://levent.ai/accelerators/smarequ/