Doubow

Multi-agent job search platform where AI drafts and humans approve.

✨ Overview

Doubow combines a modern Next.js web app with a FastAPI backend to help users discover roles, score fit, prepare applications, and safely approve outbound actions.

🧭 Discover scored jobs
🗂️ Track applications in pipeline
✅ Approve/reject drafts with HITL safety
🎯 Generate interview prep
📄 Parse resume into structured profile
💬 Unified Assistant (Messages) — chat + slash commands and structured tools with the same outcomes as Discover, Pipeline, and Approvals (/v1/agents/chat, /v1/agents/capabilities)
🌐 Job catalog ingestion — Adzuna + Greenhouse + optional Google Jobs (SerpAPI), resume-aligned preset ingest, shared jobs catalog
🤖 Monitor background agent status; optional LangGraph autopilot when enabled

🏗️ High-Level Architecture

The runtime diagram below matches docs/architecture/doubow-high-level-flow.md line-for-line; that doc adds Assistant/ingestion detail, Boundaries, observability tables, and local deployment prose. LLM is reached via domain services and the Assistant, not as API → LLM. Semantic match and offline eval are scoring-layer concerns—see At a glance and Local Validation.

flowchart LR
  U[User]
  FE["Web<br/>Next.js App Router"]
  CL["Clerk<br/>auth"]
  API["API gateway<br/>FastAPI"]
  AG["Domain services<br/>discover • score • write • apply • prep • monitor<br/>job-search pipeline"]
  ASST["Unified Assistant<br/>SSE • tools • capabilities"]
  INGEST["Job catalog ingestion<br/>Adzuna • Greenhouse • Google Jobs • resume-aligned • dedupe"]
  DB[("Supabase<br/>Postgres")]
  RD[("Redis")]
  LLM["OpenRouter<br/>tiered models"]
  OAUTH["Channels<br/>Google • LinkedIn OAuth"]
  AUTO["Autopilot<br/>runs • resume API"]
  LG["LangGraph<br/>optional parity runtime"]
  PH[("PostHog")]
  SE[("Sentry")]
  METRICS["GET /metrics<br/>Prometheus"]

  U --> FE
  FE -->|session| CL
  FE -->|Bearer JWT| API
  API -->|verify JWT| CL
  API --> AG
  API --> ASST
  API --> INGEST
  API --> OAUTH
  API --> AUTO
  API --> DB
  API --> RD
  API --> METRICS
  API --> SE
  AG --> DB
  AG --> RD
  AG --> LLM
  ASST --> LLM
  ASST --> DB
  INGEST --> DB
  AUTO --> LG
  AUTO --> DB
  AUTO --> RD
  LG --> DB
  FE --> PH
  API --> PH

At a glance

Auth & API — Dashboard in apps/web/ calls backend/api_gateway with Clerk JWT; requests are scoped by user_id with CORS tuned for local dev.
Assistant — /messages: POST /v1/agents/chat (SSE), optional ORCHESTRATOR_LLM_TOOL_ROUTING, GET /v1/agents/capabilities; tool run_job_search_pipeline matches POST /v1/agents/job-search-pipeline/run; assistant counters on GET /metrics.
Data & LLM — Supabase Postgres is durable; Redis for coordination/caches; OpenRouter tiered models for chat, drafts, prep, resume parsing, optional tool planner.
Ingestion — Adzuna, Greenhouse, optional Google Jobs (SerpAPI), resume-aligned catalog runs via catalog_ingest_orchestrator → shared jobs (dedupe, audit tables); scripts under backend/scripts/.
Scoring & outcomes — Template + semantic / lexical / LLM blend; optional per-user matching_blend_hints from stored feedback_learning; metric doubow_matching_blend_score_sync_total. Semantic match when feature-flagged; offline eval under Local Validation.
Job-search pipeline — JobSearchPipelineCoordinator runs default stages (ingest plan → profile → rescore → outbound snapshot → feedback); see Job search pipeline and docs/architecture/doubow-high-level-flow.md.
Outbound — Approvals are channel-aware (email, LinkedIn); user approval before send.
Autopilot — Background runs; optional LangGraph + checkpoints + resume API; backend/README.md for flags.
Observability — PostHog (product), Sentry (errors), Prometheus /metrics (HTTP, LLM, assistant routing/actions, matching-blend signals); optional LangChain for structured resume parsing (USE_LANGCHAIN in .env.example).

Deployment View (Local)

flowchart TB
  FE["Web<br/>localhost:3000"]
  API["api_gateway<br/>localhost:8000"]
  PG[("Supabase<br/>Postgres")]
  R[("Redis")]
  W["Workers<br/>optional"]

  FE --> API
  API --> PG
  API --> R
  W --> PG
  W --> R

Matches the local section in docs/architecture/doubow-high-level-flow.md (no OpenRouter box—traffic is HTTPS from the API/workers). Assistant, ingestion, and autopilot usually share the same API process in dev unless workers are split. Autopilot and LangGraph flags: backend/README.md.

🎯 Job search pipeline

End-to-end stages run in order via JobSearchPipelineCoordinator in the API (POST /v1/agents/job-search-pipeline/run; assistant tool run_job_search_pipeline). Implementation: backend/api_gateway/services/job_search_pipeline.py.

flowchart LR
  DC[data_collection]
  RP[resume_profile]
  JM[job_matching]
  OA[outbound_application]
  FB[feedback]
  DC --> RP --> JM --> OA --> FB

Stage	In Doubow
data_collection	Provider readiness, optional resume-aligned catalog ingest (Adzuna, Greenhouse, optional connectors), query plan from resume + preferences
resume_profile	Latest parsed résumé + preferences on file
job_matching	Recompute template-backed scores (semantic / lexical / LLM blend); optional per-user `matching_blend_hints` from stored `feedback_learning`; Prometheus `doubow_matching_blend_score_sync_total` when personalized blend differs from globals
outbound_application	Snapshot: pending approvals + application counts
feedback	Application outcomes by status + advisory `feedback_learning` snapshot; optional persist into preferences

Related UX: Discover shows fit_score / fit_reasons; the Unified Assistant can queue jobs (queue_job_to_pipeline) and run the same pipeline as the API. Approvals gate outbound send (HITL). Broader platform architecture: docs/architecture/doubow-high-level-flow.md and High-Level Architecture above.

Figures (Doubow-branded SVG)

Raster *.png (from SVG sources via rsvg-convert) so diagrams render reliably on GitHub. Editable sources: docs/architecture/job-search-pipeline/*.svg.

Assets index: docs/architecture/job-search-pipeline/README.md.

🧱 Frontend + Backend Stack

Web App (`apps/web/`)

Next.js 14 App Router + TypeScript
SWR data fetching + Zustand state
Clerk auth UI integration
PostHog client telemetry
Icons via lucide-react
Motion support via framer-motion

Backend (`backend/`)

FastAPI API gateway
SQLAlchemy + Alembic migrations
Postgres + Redis-friendly architecture
Agent/service modules for discovery, scoring, writing, apply, prep, monitor
Unified Assistant: orchestrator chat, tool executor parity with UI, optional LLM tool router
Job providers: Adzuna + Greenhouse adapters, preset ingestion, backend/scripts/*_ingestion_runner.py
Optional LangGraph autopilot runner (feature flags): parity graph nodes, graph_checkpoint on autopilot_runs, resume API for stuck runs
Optional LangChain (langchain-core) for structured resume analysis when USE_LANGCHAIN is on
PostHog-backed activation KPI endpoint

🗺️ Repo Map

apps/web/ — web application
backend/ — API, models, migrations, scripts, workers
docs/ — architecture, design system, onboarding, product maps
backend/infra/ — Docker snippets (Postgres/Redis dev stack, optional API/worker/nginx samples)

Primary references:

docs/structure.md
docs/product-panels.md
docs/architecture/doubow-high-level-flow.md
docs/architecture/daubo-architecture-claude.md
docs/architecture/daubo-design-system.md

🎨 Design, Icons, Animation

Main logo: Doubow mark (see apps/web/public/favicon.svg and apps/web/components/Logo.tsx).
Icon language: lucide-react for consistent UI semantics across dashboard and landing.
Product visual system is documented in docs/design.md and docs/architecture/daubo-design-system.md.
Core motion primitives: fade/slide progress transitions, loading indicators, and low-amplitude hover elevation.
Animation library: framer-motion (landing and interaction micro-motion).

Visual references you can embed in GitHub markdown:

docs/design-screens/target-daubo.png
docs/design-screens/landing-daubo-full-redesign.png
docs/design-screens/local-daubo-after-pixel-pass.png
apps/web/public/reference/landing/hero-dashboard-real.png

Animation guidance:

Add short .gif demos in docs/design-screens/ (Discover loading, Approvals flow, Dashboard interactions).
Link them in this README under a “Demo” section when available.

🚀 Quick Start

1) Setup environment

cp .env.example .env
cp backend/.env.example backend/.env

2) Install dependencies

npm install

3) Start web app

npm run dev:web

4) Start backend stack

docker compose -f backend/docker-compose.yml --env-file backend/.env up --build

Stop backend stack:

docker compose -f backend/docker-compose.yml --env-file backend/.env down

🧪 Local Validation

Frontend: http://localhost:3000
Auth route: http://localhost:3000/auth/sign-up
API health: http://localhost:8000/healthz (liveness), http://localhost:8000/ready (readiness — Postgres + Redis status)
Week 1 KPI snapshot: ./.venv-test/bin/python scripts/baseline_report.py
Phase 3 offline semantic precision eval: ./.venv-test/bin/python scripts/semantic_precision_eval.py --user-id <clerk_user_id>
Phase 3 with stronger outcome-proxy labels: ./.venv-test/bin/python scripts/semantic_precision_eval.py --user-id <clerk_user_id> --label-mode outcome

🗄️ Database Workflow (Supabase/Postgres)

Set DATABASE_URL in backend/.env, then migrate + seed + verify (works for local Docker and Supabase; the backend Makefile normalizes the URL for Alembic and psql):

make -C backend db-sync

Explicit steps:

make -C backend db-migrate
make -C backend db-seed
make -C backend db-verify

Reset demo fixtures only:

make -C backend db-reset-demo

📈 Telemetry And KPI (PostHog)

Frontend env:

NEXT_PUBLIC_POSTHOG_KEY
NEXT_PUBLIC_POSTHOG_HOST

Backend env:

POSTHOG_HOST
POSTHOG_PROJECT_ID
POSTHOG_PROJECT_API_KEY
POSTHOG_PERSONAL_API_KEY

Behavior:

Frontend captures product events/pageviews.
Backend mirrors product events and serves:
- GET /v1/me/telemetry/activation-kpi (avg/latest/sample size for resume -> first matches),
- GET /v1/me/telemetry/outcome-kpi (approval resolution/acceptance/send conversion rates),
- GET /v1/me/telemetry/launch-scorecard (single GO/WATCH/NO_GO snapshot combining activation, outcomes, and stability signals).

🛣️ Product Surfaces

/discover — scored job discovery + onboarding states
/pipeline — application tracking + integrity checks
/approvals — human approval gate for outbound actions
/prep — role-specific interview preparation
/resume — resume upload, parse, preferences
/messages — Unified Assistant (streaming chat, slash commands, structured account actions; capabilities from API)
/agents — redirects to /messages (bookmark compatibility)
/billing — subscription & billing (layout from docs/mockup/subscription_billing)

📚 Documentation

Job-search pipeline: section Job search pipeline above; PNG figures (+ SVG sources) under docs/architecture/job-search-pipeline/; code backend/api_gateway/services/job_search_pipeline.py
Production / staging checks for pipeline + feedback learning: docs/operations/production-job-search-verify.md
Capstone rubric + readiness (eval notes, deployment checklist, demo script, risk register): docs/capstone-scoring-sheet.md, docs/capstone-readiness.md
Architecture: high-level flow docs/architecture/doubow-high-level-flow.md; resilience (health probes, rate limits, OpenRouter circuit): docs/architecture/resilience.md
Backend detail (LLM tiers, ingestion, autopilot/LangGraph): backend/README.md
Design system: docs/architecture/daubo-design-system.md
Product panel behavior: docs/product-panels.md
Onboarding notes: docs/onboarding.md
Stack decision matrix and baseline gates: docs/stack-decisions.md

Name		Name	Last commit message	Last commit date
Latest commit History 256 Commits
.agents/skills		.agents/skills
.cursor/skills/shadcn		.cursor/skills/shadcn
.github		.github
apps		apps
backend		backend
deploy/prometheus		deploy/prometheus
docs		docs
packages/shared		packages/shared
scripts		scripts
.dockerignore		.dockerignore
.env.example		.env.example
.gitignore		.gitignore
Dockerfile		Dockerfile
README.md		README.md
fly.toml		fly.toml
package-lock.json		package-lock.json
package.json		package.json
pyproject.toml		pyproject.toml
railway.json		railway.json
skills-lock.json		skills-lock.json
turbo.json		turbo.json
uv.lock		uv.lock

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

Doubow

✨ Overview

🏗️ High-Level Architecture

Deployment View (Local)

🎯 Job search pipeline

Figures (Doubow-branded SVG)

🧱 Frontend + Backend Stack

Web App (`apps/web/`)

Backend (`backend/`)

🗺️ Repo Map

🎨 Design, Icons, Animation

🚀 Quick Start

1) Setup environment

2) Install dependencies

3) Start web app

4) Start backend stack

🧪 Local Validation

🗄️ Database Workflow (Supabase/Postgres)

📈 Telemetry And KPI (PostHog)

🛣️ Product Surfaces

📚 Documentation

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

Doubow

✨ Overview

🏗️ High-Level Architecture

Deployment View (Local)

🎯 Job search pipeline

Figures (Doubow-branded SVG)

🧱 Frontend + Backend Stack

Web App (apps/web/)

Backend (backend/)

🗺️ Repo Map

🎨 Design, Icons, Animation

🚀 Quick Start

1) Setup environment

2) Install dependencies

3) Start web app

4) Start backend stack

🧪 Local Validation

🗄️ Database Workflow (Supabase/Postgres)

📈 Telemetry And KPI (PostHog)

🛣️ Product Surfaces

📚 Documentation

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Web App (`apps/web/`)

Backend (`backend/`)

Packages