Lead AI Systems Architect (Data-Centric AI & Multimodal)

This is a remote position. We are building a an AI & Data Lab focused on social media intelligence and influencer marketing data. We are seeking a Lead AI Systems Architect to design and scale our automated classification, enrichment, and multimodal AI pipelines. You will take raw, unstructured social content (short-form video, audio, captions, and visual metadata) and convert it into high-precision structured data products.

Responsibilities Design & Build Multimodal AI Pipelines: Combine Computer Vision (scene classification, aesthetic scoring), Speech-to-Text (Whisper), and LLMs into automated ingestion workflows. Structured Data Extraction: Implement programmatic extraction frameworks (Instructor, Pydantic, DSPy, Outlines) to enforce strict, validated JSON outputs from generative models. Model Optimization & Cost Control: Transition workloads from commercial APIs to fine-tuned open-source models (Llama 3/4, Qwen, DeepSeek) running on local inference engines (vLLM, TGI, Groq).

Programmatic Labeling & Quality Control: Deploy data-centric AI frameworks (Cleanlab, Snorkel, Active Learning) to automate millions of labels and clean noisy datasets. Technical Leadership: Guide junior developers, establish engineering best practices, and set architectural roadmaps for data enrichment. Requirements 5+ years of experience in Applied AI, Natural Language Processing, or Computer Vision.

Proven experience building multimodal pipelines integrating vision, speech, and text models. Mastery of Python, PyTorch/TensorFlow, and production LLM orchestration tools ( DSPy, Instructor, Pydantic, LangChain/LlamaIndex ). Hands-on experience fine-tuning and serving open-source models using vLLM , TGI , or similar high-throughput inference engines. Experience with vector databases ( Qdrant, pgvector, Pinecone ) and RAG architectures at scale.

Preferred Qualifications Prior experience with social media data pipelines (TikTok, Instagram Reels, YouTube Shorts). Background in data-centric AI, active learning, or synthetic data generation.