Junior Software Developer — Content Parsing & Ingest

Cashmere · [Location] · [Full-time]

About us

Cashmere is the content infrastructure between the world's premium publishers and the AI economy. We work with the largest academic, scientific, professional, and trade publishers, and we build search indexes over content available nowhere else — not on the open web, not in a scrape, not in any model's training data. Demand for it is climbing far faster than we can onboard it. Every new catalog has to be received, parsed, and verified before a single query can reach it, and that pipeline is the bottleneck between a signed publisher and a live product.

The role

You work that bottleneck. It is three jobs in roughly equal measure — Python development, content-quality judgment, and project management — plus whatever else a small team needs that week.

Parse. Fix parser bugs and format edge cases, write parsers for new sources and for publisher content that doesn't fit the standard path, and tighten the quality tests that catch missing navigation, truncated sections, and dropped images before a customer does.

Review. Check parse output before it ships: structure, complete text, images, metadata. Automated tests catch a lot, but content can pass every check and still be wrong. You are the last gate.

Onboard publishers with our partnerships team. When partnerships signs a publisher, you turn "we have a deal" into "the content is in and verified." Establish exactly what we are expecting — which titles, which formats, what schedule, how many files — maintain the manifest that answers "what should we have received by now?", monitor what actually arrives, and chase every delta to closure with the publisher. This half is offline human work: reading the agreement, joining the calls, asking what nobody wrote down. It is not handed to an AI.

Resolve errors. Own the parse error queue as a working list, not a report. Triage daily, group by root cause, take it to zero.

Help everywhere else. Read our API documentation as a developer would and propose developer-experience fixes, chase slow queries and slow pages, reproduce and document bugs, make small internal-tooling changes.

What we need

• 0–2 years of professional experience, or equivalent work you can walk us through.
• Fluency writing, reviewing, and testing code with AI. This is a hard requirement. Our codebase is too large to work through unassisted and the team already works this way. Be skeptical of what comes back: you own the correctness of what you merge, and "the model wrote it" does not explain a bug.
• Working Python, basic SQL, and comfort with a terminal and Git.
• Familiarity with HTML and XML, and willingness to get deep into format specs.
• Real care about detail. Most of this job is noticing what is subtly wrong in output that looks fine at a glance.
• Project-management instincts — hold commitments across several publishers and follow up without being reminded.
• Comfort on a call with a publisher's production team, asking precise questions about their files.
• Appetite for breadth. If you want a narrow, well-fenced first job, this is not it.

Nice to have: publishing, library, or scholarly-XML background; EPUB or PDF tooling; search; query performance; technical writing.

How we work

Python shop, mainstream framework, small team — we will go through the specifics in the interview. We build with AI throughout, and a meaningful share of the pull requests you review will have been drafted by an agent. That raises the bar on judgment rather than lowering it: the scarce skill is knowing what to ask for, spotting the plausible answer that is wrong, and insisting on a test that proves it.