Meta's Llama showcase, read through its three interfaces

Three front ends, one model

Meta's January 13 post lists three organizations using Llama: WriteSea, The Washington Post, and Nanome. What's useful here isn't the shared model underneath — it's that each company shipped a visibly different interface on top of it.

WriteSea's Job Search Genius transcribes video-based mock interviews and returns performance metrics. The Washington Post's 'Ask The Post' is a text chatbot that answers in the paper's voice and links to source articles. Nanome's MARA lets scientists query a molecular structure and get answers in both text and 3D visual form. Same foundation, three completely different user surfaces.

What the interviewing tool actually renders

WriteSea's product is the most UI-heavy of the three. Per the source, it helps job seekers write custom resumes, run mock interviews, and rehearse salary negotiation — and its interface handles video-based interviews, transcribes the candidate's spoken responses, and surfaces metrics to help them improve.

That's a stack of front-end problems Llama doesn't solve by itself: capturing video, feeding audio into transcription, and presenting scored feedback in a way a nervous job seeker can act on. CEO Brandon Mitchell frames the choice as economic — open source lets them avoid API call costs and scale to over 100,000 users — but the reason those users stay is the interaction design around the model, not the model.

Transparency built into the response format

The Washington Post's design choice is the most instructive for anyone building a trustworthy front end. 'Ask The Post' answers from the paper's article archive dating back to 2016, responds in the newspaper's voice, and links to the source articles behind each answer.

That last detail is a front-end commitment as much as a data one. Rendering citations inline — and constraining the answer to published reporting — turns an open-ended chatbot into something a newsroom can stand behind. CTO Vineet Khosla puts the motivation plainly:

Open source AI is helping people stay ahead, without the limitations and cost restrictions of other proprietary AI models.Montana Labs

When the interface is a 3D scene

Nanome's MARA is the clearest case of the model serving a specialized display. Scientists view molecular structures in 3D, ask MARA questions about the structure in front of them, and get answers tied to what's on screen. The chat isn't the whole product — it's an assistant layered onto a spatial visualization tool.

Founder Steve McCloskey ties this to lowering the cost and time of drug development, work he describes as potentially taking years and billions. Nanome cites Llama's performance and adaptability, plus the collaboration it enables across universities and researchers — but the differentiator a scientist sees is a model grounded in the exact molecule they're rotating on screen.

The implication: Meta's examples reward interface work, not just model access

Read as a set, these three cases make a quiet point about open weights. Free access to Llama is the enabler each company cites for cost, but it's the surface each team built — transcribed video feedback, cited newspaper answers, a queryable 3D structure — that determines whether the tool is worth using.

For teams evaluating open source AI, the takeaway from Meta's own showcase is that the model is the starting point and the front end is where the product lives. The economic argument gets you to Llama; the interface decides what your users actually get from it.