209 Free AI Projects to Build Before 2027 (Every Link, Sorted by Level)

Published September 2026 · avivashishta.com

You commented projects, so here are all 209, with a working link for each one. They come from two open-source GitHub repos, and every project has its own folder with code and a README. I checked every link on 27 September 2026.

The two repos

Shubhamsaboo/awesome-llm-apps ~140K ★

github.com/Shubhamsaboo/awesome-llm-apps

Apache-2.0. Starter agents that run with one API key, production-style agents, multi-agent teams, RAG tutorials, MCP agents, voice agents, generative UI, memory, fine-tuning and a set of agent skills for coding agents. It was updated on 26 September 2026.

patchy631/ai-engineering-hub ~38K ★

github.com/patchy631/ai-engineering-hub

MIT. Projects sorted into beginner, intermediate and advanced: local OCR and chat apps, RAG, agent workflows, MCP servers, model comparisons, fine-tuning and full production systems.

Start with these 10

1. Chat with PDF (GPT & Llama3): The hello world of AI apps. Build it in an evening so RAG stops being a buzzword.
2. Local RAG Agent: Same idea, fully local with Llama 3.2 and Qdrant. No API key, no bill.
3. Llama OCR: Local OCR with Llama 3.2 vision. Point it at your handwritten notes.
4. GitHub MCP Agent: Your first MCP project. Ask questions about any repo in plain English.
5. Corrective RAG (CRAG): RAG that grades its own retrieval and retries. Great interview talking point.
6. Always-on Hacker News Briefing Agent: A scheduled agent that sends you a ranked daily brief. Shows you can ship something that runs without you.
7. LLM Panel Agent Team: Three models review the same diff blind, then argue. A fun multi-agent build.
8. NotebookLM Clone: RAG, citations and podcast generation in one app. A proper capstone.
9. Gemma 3 Fine-tuning: 4-bit LoRA with Unsloth. Your first fine-tune, small and readable.
10. Project Graveyard: A Claude Code / Codex skill that finds the side projects you abandoned and helps you finish one. Start here if you have a graveyard.

All 209 projects

Numbered straight through so you can say "I built #47".

Part 1: awesome-llm-apps (121 projects)

Agent Skills (8)

1. Project Graveyard: Finds every side project you abandoned, tells you why each one died, and helps you finish the one worth going back to
2. First Reader: Simulates real readers going through your draft and reports where they lose interest, where they stop reading, and what they remember afterward, without rewriting a word
3. Scope Creep Detector: Checks whether a diff grew beyond its stated intent and recommends what to keep, split, or justify
4. Commit Archaeologist: Reconstructs why a file or code region exists from its introducing commit, later edits, co-changes, and intent clues
5. Dependency Doctor: Checks a dependency manifest for standard-library pins, obsolete backports, unpinned entries, duplicate constraints, and yanked releases
6. Advisor Orchestrator Worker: Meta Loop with Claude Fable 5.1 as advisor, GPT-6 Astra as orchestrator, and Gemini 3.8 Flash as worker
7. Thinking Out Loud: Echoes a voice ramble back as a scannable brief, with the model's guesses quarantined and your reversals flagged
8. Self-Improving Agent Skills: Automatically optimize agent skills using Gemini and ADK

Starter AI Agents (13)

9. AI Blog to Podcast Agent: Turn any blog URL into a narrated podcast episode
10. AI Breakup Recovery Agent: An agent team that talks you through the post-breakup spiral
11. AI Data Analysis Agent: Ask questions of any CSV or Excel file in plain English
12. AI Medical Imaging Agent: Diagnostic analysis of X-rays and scans with Gemini
13. AI Meme Generator Agent (Browser): Makes memes by driving a real browser, not an image API
14. AI Music Generator Agent: Prompt in, MP3 track out
15. AI Travel Agent (Local & Cloud): Personalized day-by-day travel itineraries
16. AI x402 Paying Agent: An agent with a wallet that pays per-call for the data it needs: no API keys
17. Gemini Multimodal Agent: Video analysis plus web search in one agent
18. Mixture of Agents: Multiple LLMs answer, one aggregates the best response
19. xAI Finance Agent: Real-time stock analysis powered by Grok
20. OpenAI Research Agent: Multi-agent topic research with the OpenAI Agents SDK
21. Web Scraping AI Agent: Describe what to extract and the agent scrapes it

Advanced AI Agents (22)

22. AI Home Renovation Agent with Nano Banana Pro: Photos of your space in, renovation plan and photorealistic renders out
23. DevPulse AI - Multi-Agent Signal Intelligence: Aggregates and scores technical signals into a daily intelligence digest
24. AI Deep Research Agent: Comprehensive web research with the OpenAI Agents SDK and Firecrawl
25. AI VC Due Diligence Agent Team: Multi-agent startup investment analysis with Gemini 3
26. AI Research Planner & Executor (Google Interactions API): Multi-phase research with stateful conversations and auto-generated infographics
27. AI Consultant Agent: Market analysis and strategy recommendations with live web research
28. AI System Architect Agent: Architecture reviews using DeepSeek R1 reasoning plus Claude
29. AI Financial Coach Agent: Personalized budget, debt, and savings analysis
30. AI Movie Production Agent: Script drafts and casting ideas from a one-line movie concept
31. AI Investment Agent: Stock comparison reports built on Yahoo Finance data
32. Earnings Call Analyst Agent: Turns YouTube earnings calls into a playback-synced analyst workspace
33. AI Health & Fitness Agent: Tailored diet and workout plans from your goals
34. AI Product Launch Intelligence Agent: Go-to-market intelligence on competitor launches
35. AI Fraud Investigation Agent: Cross-references public records to flag facilities that don't add up
36. AI Journalist Agent: Researches, writes, and edits articles on any topic
37. AI Mental Wellbeing Agent: A coordinated agent team for mental health support plans
38. AI Meeting Agent: Context, industry insights, and strategy briefs before you walk in
39. AI Self-Evolving Agent: Agents that rewrite their own workflows with EvoAgentX
40. AI Sales Intelligence Agent Team: Generates competitive sales battle cards in real time
41. AI Social Media News and Podcast Agent: Curates your trusted sources into briefs and generated podcasts
42. Openwork - Open Browser Automation Agent (separate repo): Open-source agent that operates a real browser
43. Trust-Gated Multi-Agent Research Team: Every agent verified, every action in a hash-chained audit trail

Always-on Agents (2)

44. Always-on Hacker News Briefing Agent: A scheduled scout that ships a ranked daily brief to Slack or email
45. Release Radar Agent: Watches dependency releases and briefs you on breaking, deprecated, security, and major-version changes

Multi-agent Teams (14)

46. AI Competitor Intelligence Agent Team: Structured competitor teardowns built from their own websites
47. AI Finance Agent Team: A financial analyst team in 20 lines of Python
48. AI Game Design Agent Team: Full game concepts from a swarm of design specialists
49. AG2 Adaptive Research Team: Agent teamwork with routing and fallback, built on AG2
50. AI Legal Agent Team (Cloud & Local): Research, contract analysis, and strategy from a full legal bench
51. AI Recruitment Agent Team: Resume screening to interview scheduling, end to end
52. AI Real Estate Agent Team: Property search, market analysis, and recommendations
53. AI Services Agency (CrewAI): A digital agency that scopes and plans your software project
54. AI Teaching Agent Team: A faculty of agents that builds your complete learning path
55. Multimodal Coding Agent Team: Snap a photo of a coding problem, get a sandboxed solution
56. Multimodal Design Agent Team: Design critiques from a Gemini-powered expert panel
57. Multimodal UI/UX Feedback Agent Team: Landing page feedback plus an auto-generated improved version
58. AI Travel Planner Agent Team: A complete trip itinerary, crafted by a team
59. LLM Panel Agent Team: Three vendors review the same diff blind, then argue it out anonymously

Voice AI Agents (5)

60. AI Audio Tour Agent: Self-guided audio tours from your location, interests, and pace
61. Customer Support Voice Agent: Voice answers grounded in your own docs
62. Insurance Claim Live Agent Team: Voice claim intake on Gemini 3.8 Live that writes a field notebook, looks at damage through the webcam, and sketches the incident
63. Voice RAG Agent (OpenAI SDK): Ask your PDFs questions, hear the answers
64. OpenSource Voice Dictation Agent (Wispr Flow clone) (separate repo): Open-source dictation that types where you talk

Generative UI and Agentic Frontends (7)

65. Generative UI Starter Project: A chat-driven kanban board you and the agent work together
66. AI Financial Coach Agent: Budget, savings, and debt plans rendered as interactive cards
67. AI Dashboard Canvas Agent: Describe a dashboard in chat, charts assemble on a live canvas
68. AI MCP App Builder: Describe an MCP app, get a live sandboxed instance back
69. MCP Apps Generative UI Showcase: MCP apps that render real interactive UI, flight search included
70. AI Shadcn Component Generator: Chat your way to production-ready shadcn components
71. AI Deep Research Agent: Research where every tool call renders as a live workspace card

Autonomous Game-Playing Agents (3)

72. AI 3D Pygame Agent: DeepSeek R1 writes PyGame code, browser agents run it live
73. AI Chess Agent: Agent White vs Agent Black with validated moves
74. AI Tic-Tac-Toe Agent: Two different LLMs battle it out, move by move

MCP AI Agents (6)

75. Browser MCP Agent: Drive a real browser with natural language over MCP
76. GitHub MCP Agent: Explore and analyze any repo in plain English
77. Notion MCP Agent: Talk to your Notion pages from the terminal
78. AI Travel Planner MCP Agent: Itineraries built on live Airbnb and Google Maps data
79. Multi-MCP Agent Router: Specialist agents, each wired to its own MCP server
80. OpenAI Remote MCP Tool Bridge: Connect OpenAI function calling directly to a remote MCP server

RAG (Retrieval Augmented Generation) (21)

81. Agentic RAG with Embedding Gemma: Fully local agentic RAG with EmbeddingGemma and Llama 3.2
82. Agentic RAG with Reasoning: Watch the agent's step-by-step reasoning as it retrieves
83. AI Blog Search (RAG): Agentic search over blog content, built on LangGraph
84. Autonomous RAG: GPT-4o answers from your PDFs, falls back to web search
85. Contextual AI RAG Agent: Managed RAG: datastore to grounded chat in minutes
86. Corrective RAG (CRAG): Retrieval that grades itself and retries before answering
87. Typed Agentic RAG with Pydantic AI: Validated answers with exact citations, or a refusal when evidence is weak
88. Deepseek Local RAG Agent: Local DeepSeek reasoning over your own documents
89. Gemini Agentic RAG: Query rewriting and web fallback with Gemini Flash Thinking
90. Hybrid Search RAG (Cloud): Keyword plus vector search feeding Claude
91. Llama 3.1 Local RAG: Chat with any webpage, fully offline
92. Local Hybrid Search RAG: Hybrid search with everything running on your machine
93. Multimodal Agentic RAG: Text, PDFs, images, audio, and video, answered with citations
94. Local RAG Agent: Llama 3.2 and Qdrant, no API keys required
95. RAG-as-a-Service: A production RAG service in under 50 lines
96. RAG Agent with Cohere: Command R7B retrieval with web-search fallback
97. Basic RAG Chain: The minimal retrieval pipeline, applied to pharma research
98. RAG with Database Routing: Routes each question to the right database automatically
99. Vision RAG: Ask questions about images and PDF pages with Embed-4
100. RAG Failure Diagnostics Clinic: Find out why your RAG pipeline is wrong, systematically
101. Knowledge Graph RAG with Citations: Multi-hop answers with verifiable source attribution

AI Browser Tools (2)

102. Needle - A New Way to Find: Search webpages by meaning and highlight the strongest source sentence, using a Chrome extension powered by TypeSafe Jev
103. Ripple - Change One Thing, Find What Else Needs to Change: Find related inconsistencies and suggested fixes as you edit a Google Doc, using TypeSafe Jev and Gemini

LLM Apps with Memory (6)

104. AI ArXiv Agent with Memory: Paper search that remembers your research interests
105. AI Travel Agent with Memory: A travel assistant that remembers your preferences
106. Llama3 Stateful Chat: Session-persistent chat with Llama 3
107. LLM App with Personalized Memory: A chatbot that keeps context across conversations
108. Local ChatGPT Clone with Memory: Fully local, with a personal memory per user
109. Multi-LLM Application with Shared Memory: Different models, one shared conversation memory

Chat with X (6)

110. Chat with GitHub (GPT & Llama3): Any repo, answered in 30 lines of RAG
111. Chat with Gmail: Ask your inbox questions
112. Chat with PDF (GPT & Llama3): The classic, in 30 lines of Python
113. Chat with Research Papers (ArXiv) (GPT & Llama3): Explore arXiv conversationally with GPT-4o
114. Chat with Substack: Chat with any newsletter's archive
115. Chat with YouTube Videos: Ask videos questions via their transcripts

LLM Optimization Tools (2)

116. Toonify Token Optimization: Reduce LLM API costs by 30-60% using TOON format
117. Headroom Context Optimization: Reduce LLM API costs by 50-90%

LLM Fine-tuning (2)

118. Gemma 3 Fine-tuning: 4-bit LoRA with Unsloth, small and readable
119. Llama 3.2 Fine-tuning: Fine-tune in 30 lines, free on Colab

AI Agent Framework Crash Courses (2)

120. Google ADK Crash Course: Starter agent, structured outputs, tools (built-in, function, third-party, MCP), memory, callbacks, plugins, and multi-agent patterns. Model-agnostic.
121. OpenAI Agents SDK Crash Course: Starter agent, function calling, structured outputs, tools, memory, evaluation, handoffs, swarm orchestration, and routing logic.

Part 2: ai-engineering-hub (88 projects)

Beginner: OCR & Vision (4)

122. LaTeX OCR with Llama: Convert LaTeX equation images to code using Llama 3.2 vision
123. Llama OCR: 100% local OCR app with Llama 3.2 and Streamlit
124. Gemma-3 OCR: Local OCR with structured text extraction using Gemma-3
125. Qwen 2.5 OCR: Text extraction using Qwen 2.5 VL model

Beginner: Chat Interfaces & UI (7)

126. Local ChatGPT with DeepSeek: Mini-ChatGPT with DeepSeek-R1 and Chainlit
127. Local ChatGPT with Llama: ChatGPT clone using Llama 3.2 vision
128. Local ChatGPT with Gemma 3: Local chat interface with Gemma 3
129. DeepSeek Thinking UI: ChatGPT with visible reasoning using DeepSeek-R1
130. Qwen3 Thinking UI: Thinking UI with Qwen3:4B and Streamlit
131. GPT-OSS Thinking UI: GPT-OSS with reasoning visualization
132. Streaming AI Chatbot: Real-time AI streaming with Motia framework

Beginner: Basic RAG (6)

133. Simple RAG Workflow: Basic RAG with LlamaIndex and Ollama
134. Document Chat RAG: Chat with documents using Llama 3.3
135. Fastest RAG Stack: Fast RAG with SambaNova, LlamaIndex, and Qdrant
136. GitHub RAG: Chat with GitHub repos locally
137. ModernBERT RAG: RAG with ModernBert embeddings
138. Llama 4 RAG: RAG powered by Meta's Llama 4

Beginner: Multimodal & Media (2)

139. Image Generation with Janus-Pro: Local image generation with DeepSeek Janus-pro 7B
140. Video RAG with Gemini: Chat with videos using Gemini AI

Beginner: Other Tools (3)

141. Website to API with FireCrawl: Convert websites to APIs
142. AI News Generator: News generation with CrewAI and Cohere
143. Siamese Network: Digit similarity detection on MNIST

Intermediate: AI Agents & Workflows (13)

144. YouTube Trend Analysis: Analyze YouTube trends with CrewAI and BrightData
145. AutoGen Stock Analyst: Advanced analyst with Microsoft AutoGen
146. Agentic RAG: RAG with document search and web fallback
147. Agentic RAG with DeepSeek: Enterprise agentic RAG with GroundX
148. Book Writer Flow: Automated book writing with CrewAI
149. Content Planner Flow: Content workflow with CrewAI Flow
150. Brand Monitoring: Automated brand monitoring system
151. Hotel Booking Crew: Multi-agent hotel booking with DeepSeek-R1
152. Deploy Agentic RAG: Private Agentic RAG API with LitServe
153. Zep Memory Assistant: AI Agent with human-like memory
154. Agent with MCP Memory: Agents with Graphiti memory and Opik
155. ACP Code: Agent Communication Protocol demo
156. Motia Content Creation: Social media automation workflow

Intermediate: Voice & Audio (5)

157. Real-time Voice Bot: Conversational travel guide with AssemblyAI
158. RAG Voice Agent: Real-time RAG Voice Agent with Cartesia
159. Chat with Audios: RAG over audio files
160. Audio Analysis Toolkit: Audio analysis with AssemblyAI
161. Multilingual Meeting Notes: Auto meeting notes with language detection

Intermediate: Advanced RAG (5)

162. RAG with Dockling: RAG over Excel with IBM's Docling
163. Trustworthy RAG: RAG over complex docs with TLM
164. Fastest RAG with Milvus and Groq: Sub-15ms retrieval latency
165. Chat with Code: Chat with code using Qwen3-Coder
166. RAG SQL Router: Agent with RAG and SQL routing

Intermediate: Multimodal (3)

167. DeepSeek Multimodal RAG: MultiModal RAG with DeepSeek-Janus-Pro
168. ColiVara Website RAG: MultiModal RAG for websites
169. Multimodal RAG with AssemblyAI: Audio + vector database + CrewAI

Intermediate: MCP (Model Context Protocol) (10)

170. Cursor Linkup MCP: Custom MCP with deep web search
171. EyeLevel MCP RAG: MCP for RAG over complex docs
172. LlamaIndex MCP: Local MCP client with LlamaIndex
173. MCP Agentic RAG: MCP-powered Agentic RAG for Cursor
174. MCP Agentic RAG Firecrawl: Agentic RAG with Firecrawl
175. MCP Video RAG: Video RAG using Ragie via MCP
176. MCP Voice Agent: Voice agent with Firecrawl and Supabase
177. SDV MCP: Synthetic Data Vault orchestration
178. KitOps MCP: ML model management with KitOps
179. Stagehand × MCP-Use: Web automation with Stagehand MCP

Intermediate: Model Comparison & Evaluation (8)

180. Evaluation and Observability: E2E RAG evaluation with CometML Opik
181. Llama 4 vs DeepSeek-R1: Compare models using RAG
182. Qwen3 vs DeepSeek-R1: Model comparison with Opik
183. O3 vs Claude Code: Compare Claude 3.7 and o3
184. Sonnet4 vs O4: Code generation comparison
185. Sonnet4 vs Qwen3-Coder: Coder model comparison
186. Code Model Comparison: Frontier model code comparison
187. GPT-OSS vs Qwen3: Reasoning capabilities comparison

Advanced: Fine-tuning & Model Development (2)

188. DeepSeek Fine-tuning: Fine-tune DeepSeek with Unsloth and Ollama
189. Build Reasoning Model: Build DeepSeek-R1-like reasoning models

Advanced: Advanced Agent Systems (10)

190. Documentation Writer Flow: Agentic documentation workflow
191. Multi-Agent Deep Researcher: MCP-powered deep researcher
192. Multiplatform Deep Researcher: Multi-platform research with BrightData
193. Web Browsing Agent: Browser automation with CrewAI and Stagehand
194. Paralegal Agent Crew: Intelligent paralegal with RAG
195. FireCrawl Agent: Corrective RAG with web search fallback
196. Context Engineering Workflow: Research assistant with TensorLake and Zep
197. Parlant Conversational Agent: Compliance-driven conversational agent
198. Stock Portfolio Analysis Agent: Portfolio analysis with React frontend
199. Guidelines vs Traditional Prompt: Structured guidelines comparison

Advanced: Advanced MCP & Infrastructure (5)

200. MindsDB MCP: Unified MCP for all data sources
201. Financial Analyst DeepSeek: MCP financial analysis workflow
202. Graphiti MCP: Persistent memory with Zep's Graphiti
203. Pixeltable MCP: Unified multimodal data orchestration
204. Ultimate AI Assistant: Multi-MCP server interface

Advanced: Production Systems (2)

205. GroundX Document Pipeline: World-class document processing
206. NotebookLM Clone: Full NotebookLM with RAG, citations, and podcasts

Advanced: Advanced Agent Systems (2)

207. Build a Claude Code Harness with CrewAI: Build your own Claude Code style coding harness on top of CrewAI
208. Open Agent Builder (with Composio): Drag-and-drop workflow builder for AI agent pipelines, with Composio tools and MCP support

Advanced: Fine-tuning & Model Development (1)

209. Train YOLO26 Object Detection: Train a YOLO26 object detection model on your own custom dataset

How I'd use this

• Pick 3, not 30. One RAG app, one agent, one MCP project. That covers most AI app developer interviews.
• Don't just clone and run. Change the data source to something you care about, like your college notes or a docs site you use daily, give it a live URL and write a real README.
• Put the "what I changed" part at the top of your README. That's the part a founder reads.
• Post the build in public while you make it. A day 1 ugly demo beats a day 30 perfect demo nobody saw.
• Use Claude Code or Codex to read the code with you, but make sure you can explain every file in an interview.

A note on the count: a couple of projects share names across the two repos (there are two different "AI Deep Research Agent" builds and two "AI Financial Coach Agent" builds). Two entries in the ai-engineering-hub README pointed to folders that no longer exist, so I swapped in two real projects from the same repo that aren't in its README yet.

Want the learning path to go with the projects? Read AI Engineering from Scratch: 523 free lessons, then 25 projects to build to get hired in 2026.

Follow @avi_vashishta29 for one practical AI dev resource a day.