Overview
Archisketch is an interior design and spatial technology company building AI-powered tools for furniture discovery, floor plan search, and customer engagement. As an Applied AI Developer Intern on the Growth R&D Team, I developed end-to-end AI systems that transformed sparse, unstructured product and spatial data into scalable retrieval, recommendation, and chatbot experiences.
Work
- Developed and deployed AI-powered systems from scratch, including user persona generation agents, furniture recommendation chatbots using content-based filtering, AI-based floor plan search tools, and CRM chatbots.
- Designed and launched high-performing AI agents that reduced furniture recommendation latency from over 60 seconds to 10–18 seconds on average, while lowering cost per run from over $1.00 to approximately $0.05–$0.10.
- Achieved 92% tool and agent invocation accuracy and an 86% satisfaction rate through advanced system prompt engineering and multi-layered human plus LLM evaluation workflows.
- Built and maintained graph and vector database infrastructure using Neo4j, Chroma, and Pinecone to support context-rich retrieval and recommendation experiences.
- Implemented RAG pipelines with LlamaIndex and Cohere reranking to improve contextual relevance and user-facing AI interaction quality.
- Engineered a scalable data transformation pipeline that converted sparse and unstructured inputs, including furniture images, floor plans, and persona traits, into vector-retrieval-optimized datasets across 9 vector stores with 5k–10k items each and 3 graph databases containing 10k+ nodes and 100k+ relationships.
- Automated labeling and quality refinement for 100k+ 3D objects, improving retrieval quality and recommendation accuracy across AI-powered product features.
- Led rapid experimentation cycles under limited data and domain-expertise constraints by testing agent architectures, adapting state-of-the-art research, and delivering production-ready AI systems.
- Applied agile development practices, CI/CD workflows, and GitFlow-based collaboration to support efficient development across technical and cross-functional teams.
- Presented progress and technical outcomes to stakeholders including the CEO, CTO, and senior developers.
- Authored requirement documents, design documents, testing documentation, README files, and technical handoff materials that were approved by senior developers for clarity and completeness.
Stack/Skills
Python, LlamaIndex, Cohere Rerank, Neo4j, Chroma, Pinecone, vector databases, graph databases, RAG, AI agents, prompt engineering, recommendation systems, content-based filtering, CRM chatbots, CI/CD, GitFlow.