We are looking for an AI Engineering Intern to build a multimodal knowledge assistant that can search and answer questions across every type of company content, not just text. The goal is to let users find answers hidden in scanned documents, tables, charts, images and recordings, with clear citations they can trust.
The project involves building a solution capable of:
- Ingesting and indexing text, scanned PDFs, tables, charts, images and audio in one shared retrieval space.
- Parsing document layouts and transcribing audio to make all content searchable.
- Answering questions that combine several media types using vision-language models.
- Citing sources precisely: the exact page, image region or audio timestamp.
- Benchmarking retrieval and answer quality per media type, and tuning for speed and cost.
This internship provides hands-on experience in multimodal AI, retrieval systems and LLM evaluation.
Technical Skills
- Solid Python and experience with PyTorch or Hugging Face.
- Understanding of embeddings, vector search and RAG.
- Experience with LLMs or vision models from coursework or projects.
- Basic knowledge of APIs and Docker.
Bonus Skills (Not Required but Appreciated)
- Experience with vision-language models (Qwen-VL, ColPali).
- Knowledge of OCR and layout parsing tools (Docling, PaddleOCR).
- Familiarity with speech-to-text (Whisper).
Tech Stack
- Models: Qwen2.5-VL, ColPali, CLIP/SigLIP, Whisper
- Parsing: Docling, PaddleOCR
- Search: Qdrant or Vespa (multi-vector)
- Backend: Python, FastAPI
- Evaluation & tools: RAGAS, Docker, Git
