Skip to content

About

Geospatial job-search system: ingestion pipelines, explainable matching, application tracking, and a 3D NYC interface. Active development.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Nightshift

A geospatial job-search and application-management system for New York tech, combining job-board ingestion, explainable matching, and an interactive 3D city.

Status: active development. Public main includes M0–M5 and the initial M6 corpus expansion; M6 and deployment remain unfinished. The newer local M6 development branch is separate. See PROGRESS for acceptance evidence and outstanding limitations.

Implemented capabilities

  • Job-board adapters, background polling, canonical records and source-health tracking. Failed fetches do not silently close listings.
  • Location confidence and placement rules: city-level evidence cannot place a role on a specific building.
  • Evidence-linked matching, requirement extraction and local embeddings, with score components exposed to the reader.
  • Application tracking, interviews and user-owned records protected by authentication and ownership checks.
  • MapLibre city navigation and Three.js signal rendering, linked to a searchable list and detail panel.
  • Manual posting capture and an authenticated MCP interface: assisted captures remain proposals for user confirmation.

Architecture

Job boards → adapters → ARQ workers → canonical jobs + location evidence
                                      PostgreSQL / PostGIS / pgvector
                                                  ↕
                                  FastAPI + async SQLAlchemy
                                                  ↕
                               Next.js / React / TypeScript
                             list + MapLibre / Three.js city

Redis supports background work. Alembic manages schema changes. The architecture documents and decision records explain preservation of provenance, confidence, ownership and explicit score breakdowns.

Engineering evidence

Rendering acceptance documents a controlled synthetic case with 5,000 roles and a fixed DOM count. This is a rendering-structure check, not a production throughput claim; visual density and device coverage have documented limits.

The M5 acceptance walk checks user isolation and capture provenance, including deliberate fault injection. CI covers Python, web tests, migrations, browser acceptance and secret scanning. Historical results and failures are recorded in PROGRESS; they are not claims of tests rerun for this README.

Run locally

Requires Python 3.12+, Node 20+, and Docker Desktop or OrbStack.

git clone https://github.com/Tahmudun/Nightshift.git
cd Nightshift
make doctor
make setup
make demo

Open http://localhost:3000. The offline demo uses a labelled committed job-board fixture. Outbound HTTP is disabled by default. For live ingestion, review the board registry and configuration, set OUTBOUND_HTTP_ENABLED=true in your local .env, then run make ingest. Never commit credentials or personal job-search data.

make check           # format, lint, typecheck and unit/integration tests
make test-e2e        # browser tests

Test database warning: the Python suite can truncate the configured development database. Use a disposable test database; preserve personal application data before running it. See PROGRESS and CLAUDE.md.

Stack and repository

Python, FastAPI, async SQLAlchemy, PostgreSQL, PostGIS, pgvector, Redis, ARQ, Alembic; TypeScript, Next.js, React, MapLibre, Three.js; pytest, Vitest, Playwright, mypy and GitHub Actions.

apps/web/ contains the frontend; services/api/ contains API and workers; infra/ contains local infrastructure; data/ contains the reviewed board registry. Existing design, review and evaluation documents remain under docs/.

Remaining work

M6 data and geographic coverage, later visual refinement, deployment and operational qualification remain in progress. No working public deployment is claimed here. Offline examples are labelled; historical live-ingestion counts describe a dated corpus rather than current users or production scale.

Development approach

Designed and developed by Tahmudun Nabi using Claude Code and Codex assistance, with responsibility for product decisions, architecture, debugging, testing, integration and iteration. Existing co-author attribution is preserved.

About

Geospatial job-search system: ingestion pipelines, explainable matching, application tracking, and a 3D NYC interface. Active development.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages