Selected work / 01

Job Search OS

A private operating system for discovering roles, reviewing fit, assembling evidence-backed application materials, and tracking the work around a job search.

My role: Independent product engineer
Scope: Product strategy, interface design, and full-stack implementation
Stage: Protected single-user production app
Core technologies: Next.js · TypeScript · LangGraph · LangSmith

Job Search OS dashboard showing the daily review cockpit and application pipeline

Case study

Turning a fragmented search into one operating system.

The problem

A serious job search splinters across browser tabs, spreadsheets, drafts, inboxes, and repeated research. The difficult part is not producing more material. It is knowing what is credible, what needs review, and what can move forward.

What I built

I designed and built a single-user product that joins role discovery, fit review, career evidence, application materials, tracking, email operations, and LinkedIn planning. LangGraph gives the application assistant and recruiting agency durable workflow state with Postgres checkpointing. LangSmith adds optional redacted tracing for agent and workflow diagnostics. Each workflow has visible state, and consequential actions stay in my hands.

Key decisions

01Review before action
The system can prepare work, but submissions and other external actions stop at an explicit approval gate.
02Evidence over invention
Career claims stay tied to source material, with structured outputs and deterministic fallbacks when model output is incomplete.
03Make orchestration observable
LangGraph owns pause, resume, retry, and checkpoint state where durable orchestration is useful. LangSmith redacted tracing and local quality records help diagnose failures without exposing application materials by default.

Inside the work / 07 views

A working product, end to end

The system spans the daily cockpit, guarded application execution, live search diagnostics, human review, resume operations, and reusable search strategy.

Job Search OS Apply Sprint interface showing readiness, blocked claims, and the next application workflow — 02
**Application assistant**
A review-led sprint surface that prepares application work while keeping final submission with the user.

Job Search OS Search Operations interface with live run metrics and diagnostic charts — 03
**Search operations**
Live discovery telemetry makes yield, filtering, blockers, and handoff state visible while a search is running.

Job Search OS Agent Review Board showing material warnings, agent activity, and prioritized recommendations — 04
**Agent review board**
Recommendations, warnings, and material checks meet in one place before they become decisions.

Resume workspace

Resume source material, parsed career details, and role-specific documents live in one connected workspace.

Search profiles

Reusable search strategies make target roles, filters, profile health, and expansion opportunities visible before a run begins.

Job Search OS Company Sources interface showing discovery controls and a curated source roadmap — 07
**Company sources**
A curated source registry keeps direct company pages, ATS feeds, marketplaces, and search adapters legible and controllable.

Technical view

Architecture

Human-reviewed career operations13 components · 13 connections

Four product surfaces share one evidence and orchestration layer; every consequential external action stops at an approval boundary.

Product surfacesWays Carl enters and reviews work

Application + orchestrationServer-owned workflow boundary

Durable stateShared operational record

Controlled exitsNo autonomous submission

Next.js · React
Daily cockpit
Search, fit, materials, review
Chrome extension
Browser capture
Role and source intake
Slack
Jolene command
Private operator channel
MCP · stdio
Local tool access
Tracking and preparation tools
Next.js App Router
Workflow API
Typed routes and policy checks
OpenAI · LangSmith
Specialist agents
Structured outputs + redacted traces
pgvector
Evidence RAG
Approved career evidence only
LangGraph
Durable workflows
Postgres checkpointing + review state
Node.js · Redis
Background worker
Embeddings and queued work
PostgreSQL
System of record
Roles, claims, runs, materials
pgvector
Semantic index
Career evidence embeddings
Policy gate
Owner approval
Review before consequence
Manual / approved
External channels
ATS, email, LinkedIn

surfaceservicedataaiintegrationcontrolruntime

Read system connections

Daily cockpit → Workflow API
Browser capture → Workflow APIcapture
Jolene command → Specialist agents
Local tool access → Workflow API
Workflow API → Specialist agentsplan
Workflow API → Evidence RAG
Specialist agents → Durable workflows
Evidence RAG → Semantic index
Durable workflows → System of record
Durable workflows → Background worker
Background worker → System of record
Specialist agents → Owner approvalpropose
Owner approval → External channelsapproved only

Built so far

Combines job discovery, fit review, application materials, tracking, email operations, and LinkedIn content workflows in one product, with LangGraph and Postgres checkpointing for durable assistant and recruiting-agency workflows plus optional redacted LangSmith tracing.
Uses structured model outputs, source-backed career evidence, and deterministic fallbacks where model output is not sufficient.
Keeps submissions and other consequential external actions behind explicit human approval.

Where it stands

This is a production application built for one person rather than a public applicant service.
Automated recommendations support review; they do not replace the owner’s judgment or approval.

Repository

Continue exploring

Next case study

View all selected work

Geospatial / Aviation intelligence

Flight Tracker AI

Deployed read-only portfolio demo

Open case study