Compelo
The Platform
A multi-agent architecture that compresses a week-long candidate evaluation cycle into under 5 minutes.
/ 7 SPECIALIZED AGENTS
Seven purpose-built agents handle candidate discovery, fit scoring, and retrieval-augmented generation across resume and interview data. Each agent owns a single responsibility—sourcing, parsing, ranking, summarizing, alerting, scheduling, and reporting—coordinated through an orchestration layer that ensures deterministic, auditable outputs.
BEFORE
Week-long manual screening cycles with inconsistent evaluation criteria across recruiters.
AFTER
Automated end-to-end pipeline delivering ranked, scored candidates with cited evidence in under 5 minutes.
RAG
Retrieval-augmented generation across resumes and interview transcripts surfaces relevant experience that keyword search misses.
The CRM
An ATS-style CRM managing 1,000+ candidates across 5-stage hiring lifecycles.
Kanban Pipeline
Drag-and-drop board across 5 hiring stages with real-time status sync.
SLA Alerting
Automated alerts when candidates stall in a stage beyond configurable thresholds.
RBAC
Role-based access control scoping visibility by team, role, and hiring manager.
Bulk Import
CSV and API-based candidate ingestion with deduplication and validation.
Lifecycle Tracking
Full audit trail from sourced → screened → interviewed → offered → hired.
1,000+ Candidates
Production-tested at scale across multiple concurrent hiring pipelines.
Engineering
Score Calibration
Built a dynamic calibration system that corrects LLM scoring bias of ±8 points. The system continuously benchmarks agent outputs against recruiter-labeled ground truth, achieving 95.7% agreement with human evaluators. Calibration weights update per-agent and per-role to account for domain-specific scoring drift.
REST API
Designed and shipped a 26-endpoint Dockerized microservice. Secured with HMAC webhook verification, API-key authentication, and per-endpoint rate limiting. The service handles candidate CRUD, agent orchestration, pipeline management, and reporting—all behind a versioned API contract.
Testing
2,000+ unit tests across 105 files using Vitest, plus Playwright end-to-end suites covering critical hiring flows. Maintained 90%+ code coverage with CI gating—no PR merges below the threshold. Test infrastructure includes factory fixtures, mock agent responses, and deterministic seed data.
END OF FILE
Return