System / 002
JASPI
AI hiring infrastructure
Role
Backend & Cloud Engineering
Period
Apr 2025 – Present
Domains
3
Context
Power Match (Denmark)
Primary Outcome
Production backend powering AI screening of hundreds of applicants simultaneously with zero-downtime deploys.
Last updated 2026-08-11
01 / Context
Jaspi is an AI-powered hiring platform that automates the full recruitment lifecycle — from job post generation and candidate sourcing to async AI video interviews and real-time fit scoring. Companies use Jaspi to hire at a flat $200/job rate, eliminating the need for traditional recruiters.
02 / Problem
The backend needed to handle AI screening pipelines, async chat and video interviews, real-time candidate scoring, and third-party integrations, all reliably at production scale — with no room for downtime during live hiring workflows.
04 / Architecture
CLIENT | v API GATEWAY | v HIRING SERVICE ----> AI SCREENING PIPELINE | | v v POSTGRESQL AI / ML PROVIDERS DOCKER + CI/CD ----> AWS (ZERO-DOWNTIME DEPLOY)
The hiring service owns job and candidate state in PostgreSQL and hands off to an AI screening pipeline for scoring and interview evaluation, so a slow AI provider response degrades scoring latency without taking down core hiring operations. Deployment runs through Docker and CI/CD on AWS specifically to support zero-downtime releases during active hiring cycles.
05 / Engineering Decisions
Decision / 01
AI screening needs to run for hundreds of applicants concurrently without becoming a bottleneck for the rest of the platform.
Chosen: AI screening pipelines built as a separate service boundary from core hiring APIs.
Why: Keeps AI provider latency and failure modes isolated from job posting, candidate sourcing, and account management.
Trade-off: More moving parts to deploy and monitor than a single monolithic service.
Decision / 02
The platform cannot go down during active hiring workflows — candidates and companies are mid-interview at any given time.
Chosen: Docker-based deployments with CI/CD pipelines built for zero-downtime releases.
Why: Shipping is frequent in a fast-moving product; deploys can't be a source of risk.
Trade-off: Requires more deployment discipline (health checks, rollout gating) than a simpler restart-based deploy.
08 / Technology
BACKEND
NestJS, TypeScript, Node.js
DATA
PostgreSQL, Redis
INFRASTRUCTURE
AWS, Docker, CI/CD
07 / Contribution
My Contribution
- —Designed and maintained backend services powering AI-driven hiring flows — job post generation, candidate sourcing, and screening pipelines
- —Built secure, scalable REST APIs supporting async video interviews, AI fit scoring, and real-time candidate ranking
- —Architected cloud infrastructure on AWS with Docker-based deployments and CI/CD pipelines for zero-downtime releases
- —Integrated AI/ML services into the hiring workflow, working closely with product and ML teams
- —Improved infrastructure reliability through monitoring, alerting, and automated recovery
09 / Outcome
- —Production-grade backend powering the 48-hour hiring promise
- —AI screening of hundreds of applicants simultaneously
- —Zero-downtime deployment model maintained through active development
10 / Lessons
- Isolating AI provider calls behind a dedicated service boundary paid off the first time a provider had a latency spike — the rest of the platform stayed unaffected.
- Zero-downtime deploys matter most exactly when you're shipping fastest — early investment in the release pipeline kept velocity high without adding production risk.