A playable archive of 30 production-shaped CrewAI mini-projects. Beginner → Intermediate → Advanced. Each one ships a Python pipeline and a runnable web demo.
Every demo on the site runs against hardcoded fixtures by default — zero API credits consumed. Each project also defines an optional endpoint so the same UI can be flipped to a real FastAPI backend with NEXT_PUBLIC_USE_REAL_BACKEND=true.
|
10 single-agent pipelines Foundational CrewAI patterns: one agent, one task, one clean output. The fastest way to get an LLM doing something useful.
|
10 multi-agent crews Sequential and parallel crews where 2-4 agents pass structured context: research → draft → review patterns.
|
10 production pipelines Real tool integrations, parallel orchestration, and FastAPI surfaces. Built like things you'd actually deploy.
|
| # | Agent | What it does | Demo |
|---|---|---|---|
| 01 | Resume Screener | Scores a candidate resume against a job description and explains its reasoning. | Run |
| 02 | Review Sentiment | Classifies a customer review into positive, negative, or mixed and returns the top theme. | Run |
| 03 | Email Tone Rewriter | Rewrites a draft email in a target tone while preserving the underlying ask. | Run |
| 04 | Meeting Notes | Turns a raw transcript into a tight summary, decisions list, and tagged action items. | Run |
| 05 | Cold Outreach | Generates a short, personalized cold email tying a prospect's role to your value prop. | Run |
| 06 | Code Explainer | Explains a code snippet in plain English with intent, control flow, and gotchas. | Run |
| 07 | Social Caption | Three platform-tailored captions for X, LinkedIn, and Instagram from one idea. | Run |
| 08 | Legal Simplifier | Translates a contract clause into plain English and surfaces the top three risks. | Run |
| 09 | Interview Questions | Generates a balanced set of behavioral, technical, and situational interview questions. | Run |
| 10 | Startup Validator | Pressure-tests a startup idea and gives a candid go / no-go score. | Run |
| # | Agent | What it does | Demo |
|---|---|---|---|
| 01 | SEO Blog Pipeline | Three-agent crew that researches keywords, drafts a post, and SEO-optimizes the copy. | Run |
| 02 | Support Ticket Resolver | Triages a ticket, drafts a customer reply, and decides whether to escalate. | Run |
| 03 | Competitor Intel | Crawls a competitor's site and posts to produce a positioning + pricing brief. | Run |
| 04 | Literature Reviewer | Searches recent papers, clusters findings, and produces a synthesized review. | Run |
| 05 | Job Application Assistant | Tailors a resume to a JD and drafts a matching cover letter with citations. | Run |
| 06 | Dependency Vuln Scanner | Scans a manifest for CVEs and proposes the safest upgrade path. | Run |
| 07 | Launch Checklist | Builds a launch-week checklist tailored to a product's stage and channels. | Run |
| 08 | Podcast Show Notes | Turns a transcript into structured show notes with chapters and quotables. | Run |
| 09 | Real Estate Analyzer | Pulls comps for an address and produces a price-band estimate with confidence. | Run |
| 10 | Incident Postmortem | Synthesizes incident logs into a blameless postmortem with timeline and 5-whys. | Run |
| # | Agent | What it does | Demo |
|---|---|---|---|
| 01 | Hiring Pipeline | Ingests a resume, builds a profile, matches it to live roles, drafts outreach. | Run |
| 02 | News Fact-Checker | Decomposes a claim into atomic assertions and rates each with citations. | Run |
| 03 | Price Intelligence | Tracks a product's price across competitors and recommends a positioning move. | Run |
| 04 | Crypto Portfolio Advisor | Scores concentration risk and recommends rebalancing moves with target weights. | Run |
| 05 | Repo Health Auditor | Audits a GitHub repo across velocity, CI reliability, and review hygiene. | Run |
| 06 | Travel Planner | Builds a multi-day itinerary with weather-aware day plans. | Run |
| 07 | Account Sales Intel | Builds a sales-ready brief on a target account with buying triggers. | Run |
| 08 | License Auditor | Walks a dependency tree and produces a counsel-ready license remediation plan. | Run |
| 09 | Content Repurposer | One long-form post → tweet thread + LinkedIn essay + 60-second video script. | Run |
| 10 | Energy Optimizer | Reads building usage + tariffs and prescribes load-shifting to lower the bill. | Run |
outskill-mini-projects/
├── projects/ # Python CrewAI pipelines (the actual agents)
│ ├── beginner/ # 01..10 single-agent
│ ├── intermediate/ # 01..10 multi-agent crews
│ ├── advanced/ # 01..10 production pipelines (some w/ FastAPI)
│ └── pyproject.toml # uv-managed workspace
│
├── web/ # Next.js 15 + React 19 demo site
│ ├── app/ # App Router — home, /[tier], /[tier]/[slug]
│ ├── components/ # AppShell, Terminal/*, ProjectCard, DemoPlayground
│ └── lib/projects/ # Per-project fixtures + the runProject() adapter
│
└── design/stitch/ # Source mockups (Stitch "AGENT_ARCHIVE" design system)
Backend (Python)
crewai— multi-agent orchestrationpydantic— typed inputs / outputsfastapi+uvicorn— production projects expose anapi_server.pyOpenRouter,EXA,Adzuna,Open-Meteo,CoinGecko— live tool integrations
Frontend (TypeScript)
next@15.5(App Router, RSC) onreact@19tailwindcss@3.4with a custom Stitch-inspired token system (night,terminal,volt,cyan,magenta)lucide-reactfor icons- Space Grotesk · Inter · Fira Code, loaded via
next/font
cd web
npm install
npm run dev # http://localhost:3000The site runs entirely on hardcoded fixtures by default — no API keys needed.
cd projects
uv sync # installs the workspace
uv run python -m beginner.01_resume_screener.mainEach Python project reads its API keys from environment variables (e.g. OPENROUTER_API_KEY, EXA_API_KEY). Copy them into a root-level .env (already gitignored).
Every web demo defines a ProjectDef with an optional endpoint. The shared adapter in web/lib/projects/runner.ts decides at runtime:
if (process.env.NEXT_PUBLIC_USE_REAL_BACKEND === "true" && def.endpoint) {
return fetch(def.endpoint, { method: "POST", body: JSON.stringify(input) });
}
return streamFromFixture(def);So shipping a real agent is two steps:
- Run the matching
api_server.pyfromprojects/<tier>/<n>_<name>/. - Set
endpointon theProjectDefand flip the env flag.
Same UI, real outputs.
The web app is deployed to Vercel:
- Production: web-blue-sigma-80.vercel.app
- Project:
ishan-duttas-projects-a63436c2/web - Framework: Next.js (auto-detected)
- Region:
iad1
Re-deploy:
cd web
vercel --prodBuilt with care for the Outskill GenAI Mastery Challenge.