The Problem
Building production-grade LLM applications requires assembling many moving parts: agent orchestration, retrieval pipelines, web scraping, memory, and model access. Most teams start from scratch, reinventing the same patterns and debugging the same integration issues. This repo compresses that learning curve by providing a catalog of working examples across the AI application spectrum.
What This Does
awesome-ai-apps is a curated collection of 70+ LLM-powered application examples, organized by complexity. The advanceaiagents/ directory contains the bulk of the material—186 files spanning 13 distinct projects. These range from a candidateanalyser (resume screening with hiringprompts.yaml) to an ai-hedgefund (a TypeScript-based multi-agent trading system with parallel analyst steps) to a conferenceagnositccfpgenerator (a full RAG pipeline with scrapers, vector search, and a research agent).
The projects share common infrastructure patterns. Most use pyproject.toml for dependency management, which implies pip or uv installation. Several include Docker support (ai-hedgefund/Dockerfile, financeserviceagent/Dockerfile) and CI via GitHub Actions (.github/workflows/lint.yml). The README positions this as a learning resource, not a production library—each project is self-contained and meant to be studied or adapted.
How To Use It
The repo is structured as independent projects, so there's no single build command. Each subdirectory under advanceaiagents/ is its own application.
Setup: For Python projects like candidateanalyser, install dependencies from requirements.txt or pyproject.toml:
cd advanceaiagents/candidateanalyser pip install -r requirements.txt # or: pip install -e .
For the TypeScript ai-hedgefund, use bun (a bun.lock is present) or npm with package.json.
Configuration: Most projects require API keys. Look for .env.example files (e.g., conferenceagnositccfpgenerator/.env.example, jobfinderagent/.env.example). Copy these to .env and populate with your keys. Some projects use api.env (e.g., carfinderagent/api.env).
Running it: Entry points vary. Some are CLI scripts (candidateanalyser/main.py), others are web apps (contentteamagent/app.py, smartgtmagent/app.py). The ai-hedgefund uses a Motia workflow runner (motia-workbench.json).
python main.py # candidateanalyser python app.py # contentteamagent, carfinderagent
The README documents per-project instructions; check each subdirectory's README.md for specifics.
Real-World Use
A practical scenario: a team building a job-search assistant could study jobfinderagent/ (which uses MCP server patterns and Bright Data for scraping) and candidateanalyser/ (which screens resumes via a YAML-configured prompt pipeline). These two projects together demonstrate the full loop—sourcing candidates and evaluating them—without requiring the team to design the integration points from first principles.
Code Health & Issues
Medium - Missing lockfiles: ai-hedgefund/package.json has no corresponding package-lock.json (only bun.lock), so npm install builds are non-reproducible. The Python projects rely on requirements.txt without pinned hashes. Low - Committed secrets: carfinderagent/api.env and smartgtmagent/api.env are committed to the repo. These may contain placeholder keys, but the pattern is risky. Low - Minimal test coverage: Only 1 test file exists across 200 files. This is a learning resource, so tests are secondary, but production adoption would require adding them. Low - Data files in repo: pricemonitoringagent/productdata.json and tracked_urls.json are runtime state committed to source control. Low - Mixed dependency ecosystems: The repo spans Python, TypeScript, and Docker, which is fine for a catalog but means no unified build or dependency story.
The Bottom Line
This is a strong reference catalog for teams learning to build LLM applications. The examples cover real patterns—RAG, multi-agent workflows, MCP servers—and each project is self-contained enough to run independently. It's not a production framework; you'd use it to learn patterns and borrow code, not as a dependency. Teams starting their first AI application will find it more useful than those with established infrastructure.