How LaComplice turns influencer photos into shoppable fashion with AI

LaComplice is a fashion technology company incubated at Harvard Innovation Labs. It develops B2B solutions that help leading fashion, retail and luxury brands turn creator content into a product discovery and shopping channel.

Its clients operate large, frequently changing catalogues and expect every digital touchpoint to meet the standards of a global consumer brand. Visual matches therefore had to be accurate enough for LaComplice to use the product in conversations and live projects with some of the most recognised names in fashion.

Influencer content creates demand without carrying the information required to convert it into a purchase. A photograph contains no SKU, product name or catalogue reference. The system first has to separate the individual garments inside a complete outfit and then find their closest counterparts among thousands of products.

This was 2024. General-purpose vision models could already describe an image or recognise broad categories such as a dress or handbag. Reliable catalogue matching was a different problem. Lighting, poses, overlapping garments, partial visibility and seasonal variations could all change how the same type of product appeared.

LaComplice wanted that matching to work in both directions. Someone viewing an influencer photo should be able to discover similar products available from the brand. A brand should also be able to select an item and find creator content featuring the same or a visually similar product.

Every result carried reputational weight. When the product is being presented to a top-tier fashion or luxury brand, a handful of convincing examples cannot compensate for inconsistent matching across the rest of the catalogue.

Tailor developed an AI engine that interprets fashion imagery and connects it with product inventory through visual similarity.

The system combines:

• Image segmentation to isolate individual garments and accessories from photographs containing complete outfits.

• Visual analysis using OpenAI models to extract the characteristics of each item.

• Vector search in Chroma to find catalogue products with similar visual attributes, even when the image contains no searchable text or product reference.

• Bidirectional matching between influencer content and brand inventory.

• Human review so LaComplice can validate results, correct uncertain matches and improve the quality of the experience.

LangChain orchestrates the flow between image analysis, segmentation, vector search and review.

The first matching engine gave LaComplice a working technical foundation for taking its product vision to fashion brands. It also opened a longer collaboration with Tailor, expanding from visual matching into image modification and further AI use cases.

The human-review layer allowed the team to combine automated processing with the level of control required for brand-facing results. LaComplice could refine uncertain matches without returning to a completely manual workflow.