Technical description

Node 20+, zero dependencies. Data lives in JSON/JSONL files; the program ships a CLI and a small HTTP server. Fit score and application packs are computed deterministically from the profile and the ad text — no model, no network.

How to run it

node src/cli.mjs serve                  # HTTP server (PORT, default 8795; this demo runs on 8797)
node _qa/smoke.mjs                      # 9/9 checks
PROFILE_FILE=data/profile.demo.json node src/cli.mjs serve   # demo candidate

# CLI, bez servera
node src/cli.mjs jobs
node src/cli.mjs pack --id <jobId>

API

method	path	what it does
`GET`	`/health`	liveness + job count
`GET`	`/metrics`	pipeline counters and response rate
`GET`	`/jobs`	all ads with fit scores
`GET`	`/jobs/:id`	a single ad
`GET`	`/jobs/:id/pack`	the finished application pack
`POST`	`/jobs/:id/apply`	mark as applied
`POST`	`/jobs/:id/status`	move status (reply, interview…)
`POST`	`/import`	import a new ad from text

Data

data/jobs.json — ads (company, title, pay, requirements, link)
data/profile.json — the candidate (skills, claims, roleFamilies); the demo uses data/profile.demo.json
data/pipeline.jsonl — application pipeline, append-only

Limits — what this demo does not do

No database — JSON files (enough for one candidate; a team needs Postgres).
No authentication: anyone with the URL can change status. The demo is deliberately open.
It does not send applications — no mail integration; the pack is copied by hand.
It does not scrape job boards — an ad is imported as text (POST /import).

How it is deployed

One Node process behind nginx, a systemd unit, Let’s Encrypt. The recipe is in E:\poslovi\DEMO-REGISTAR.md — the same for every program.

The demo cannot spend money: no AI keys and no outbound calls.