5 Must‑Try Open‑Source AI Projects You Can Start Using Today

1. Large‑Model Assisted PPT Generation

AiPPT ( https://github.com/veasion/AiPPT) is an open‑source project that generates PowerPoint presentations from natural‑language prompts. It provides three core capabilities: AI‑generated slides, parsing PPT files into JSON, and rendering JSON back to PPT.

Demo output is available at http://www.veasion.cn/AiPPT/.

2. Open‑Source LLM Application Development Platform

Dify ( https://github.com/langgenius/dify) offers a visual canvas for building AI workflows, including agents, Retrieval‑Augmented Generation (RAG) pipelines, model management, and observability.

Workflow Builder : drag‑and‑drop interface to construct and test AI workflows.

Comprehensive Model Support : integration with dozens of inference providers and hundreds of proprietary/open‑source LLMs such as GPT, Mistral, Llama 3, and any OpenAI‑compatible API. Full provider list at https://docs.dify.ai/getting-started/readme/model-providers.

Prompt IDE : interface for crafting prompts, comparing model performance, and adding text‑to‑speech.

RAG Pipeline : built‑in support for extracting and retrieving text from PDFs, PPTs, and other common document formats.

Agent Functions : define agents via LLM function calling or ReAct; includes 50+ built‑in tools (Google Search, DALL·E, Stable Diffusion, WolframAlpha).

LLMOps : monitoring and analysis of application logs and performance for iterative improvement.

Backend‑as‑a‑Service : all features exposed via APIs.

3. Run AI on Any Infrastructure

SkyPilot ( https://github.com/skypilot-org/skypilot) enables execution of AI and batch jobs on Kubernetes or 14 cloud providers, including AWS, GCP, Azure, OCI, Lambda Cloud, Fluidstack, RunPod, Cudo, Digital Ocean, Paperspace, Cloudflare, Samsung, IBM, Vast.ai, VMware vSphere, and Nebius. Supported infrastructures are listed at

https://docs.skypilot.co/en/latest/getting-started/installation.html

.

Example command to launch the 671B DeepSeek‑R1 model:

$ sky launch -c r1 llm/deepseek-r1/deepseek-r1-671B.yaml --retry-until-up

Relevant configuration file (llm/deepseek-r1/deepseek-r1-671B.yaml):

name: deepseek-r1

resources:
  accelerators: {H200:8, H100:8}
  disk_size: 1024  # Large disk for model weights
  disk_tier: best
  ports: 30000
  any_of:
    - use_spot: true
    - use_spot: false

num_nodes: 2  # Number of nodes; may vary with accelerators

setup: |
  uv pip install "sglang[all]>=0.4.2.post4" --find-links https://flashinfer.ai/whl/cu124/torch2.5/flashinfer
  sudo bash -c "echo 'vm.max_map_count=655300' >> /etc/sysctl.conf"
  sudo sysctl -p

run: |
  MASTER_ADDR=$(echo "$SKYPILOT_NODE_IPS" | head -n1)
  TP=$(( $SKYPILOT_NUM_GPUS_PER_NODE * $SKYPILOT_NUM_NODES ))
  python -m sglang.launch_server \
    --model deepseek-ai/DeepSeek-R1 \
    --tp $TP \
    --dist-init-addr ${MASTER_ADDR}:5000 \
    --nnodes $SKYPILOT_NUM_NODES \
    --node-rank $SKYPILOT_NODE_RANK \
    --trust-remote-code \
    --enable-dp-attention \
    --enable-torch-compile \
    --torch-compile-max-bs 8 \
    --host 0.0.0.0 \
    --port 30000

4. Curated Collection of LLM Applications

The repository “awesome‑llm‑apps” ( https://github.com/Shubhamsaboo/awesome-llm-apps) aggregates practical LLM applications built with Retrieval‑Augmented Generation and AI agents. It includes examples using OpenAI, Anthropic, Google models as well as open‑source models such as DeepSeek, Qwen, and Llama, all runnable locally.

5. One‑Click HD Short‑Video Generation

MoneyPrinterTurbo ( https://github.com/harry0703/MoneyPrinterTurbo) generates high‑definition short videos automatically. Providing a video theme or keywords produces a script, visual assets, subtitles, background music, and composes the final video.