The company is building open research agents that search the web and other sources, then return answers with citations that can be inspected.
It is an open-source research platform that lays out information as a self-organizing map. Users search web and academic sources, pull quotes and citations from articles, PDFs and videos, arrange findings in shared editors, and use agents to draft transparent outlines and answers.
The platform is live. The company is now approaching venture capital firms, accelerators and grant programs for funding.
This is a pre-funding, early-stage role paid only in equity, up to 5 to 10%, set by scope, seniority, expected commitment, vesting and the final agreement.
If funding closes, the target salary is $150,000 a year. It is not guaranteed and depends on fundraising, board and company approval, business conditions and a signed employment agreement.
You would help shape the product, the technical direction, the fundraising story and the early company foundations.
This posting does not itself create employment, pay or a binding offer. Final terms require written agreements.
Responsibilities
Design, build, test and run production AI systems, from prototype to deployment.
Build retrieval and deep-research pipelines that combine web search, content extraction, ranking, planning, tool use, memory and answer generation.
Extract clean structured content and metadata from hard sources such as scanned or complex PDFs, JavaScript-heavy pages, long articles, citations, tables and video transcripts.
Improve source ranking using retrieval models, quality heuristics, domain reputation, metadata, freshness, user intent and learned rerankers.
Build LLM orchestration that works with several providers and open-weight Llama-family models.
Create routing policies that pick the right model and workflow for each task while controlling cost, protecting quality and leaving users in control.
Create offline and online evaluation sets, test suites and dashboards for agent behavior.
Set clear standards so the system separates sourced facts from inference, shows provenance, states uncertainty and never presents unsupported claims as evidence.
Build embedding, vector, hybrid search, clustering, topic modeling and graph systems for research collections that grow over time.
Work with product and full-stack engineers to turn ML capabilities into fast, understandable interfaces.
Contribute production code to a TypeScript-first monorepo and publish reusable open-source packages where it makes sense.
Write technical documentation, reproducible benchmarks and implementation notes for contributors and users.
Take part in architecture decisions across Cloudflare Workers, serverless APIs, queues, durable state, object storage, databases, browser rendering and optional GPU or container workloads.
In the first 30 days, learn the codebase, local workflow, API surface and existing agent packages, map the research pipeline from query to final response, set baseline evaluation cases, and ship one fix to a high-impact quality issue.
In days 31 to 60, own a major AI track such as hybrid retrieval, research-agent planning, citation verification, document intelligence or topic-map generation, build a measurable evaluation loop, and expose AI features through a clear workflow instead of a chat box.
By the first 90 days, launch a substantial feature or infrastructure improvement, document the architecture, metrics, limits and operations, and help set the next quarter's AI roadmap from measured user needs.
Requirements
Strong experience building and shipping production software.
Professional or substantial project experience with LLM applications, NLP, information retrieval, search, RAG, document intelligence or agent workflows.
Strong TypeScript or JavaScript skills and comfort working in a modern monorepo.
Practical Python for experiments, data pipelines, evaluation, model tooling or GPU services.
Experience calling LLM APIs and/or running open-weight models locally or on hosted infrastructure.
Familiarity with embeddings, vector databases or indexes, retrieval pipelines, reranking and chunking.
Experience designing systems that handle messy, unstructured, real-world content.
Good judgment about reliability: you test assumptions, measure quality, instrument systems and think about failure modes.
Ability to explain complex AI system behavior to engineers, contributors and non-technical users.
Comfort in a fast-moving startup and open-source setting where requirements change as users show what matters.
A portfolio, GitHub profile, shipped product, research project, open-source contribution or technical write-up that shows relevant work.
Preferred: experience building web search, deep-research or citation-answering systems.
Preferred: experience with retrieval evaluation, ranking metrics, query relevance or search-quality analysis.
Preferred: experience with agent frameworks and protocols such as Vercel AI SDK, Mastra, MCP, LangChain, LangGraph or custom orchestration.
Preferred: experience deploying on Cloudflare Workers, Durable Objects, Queues, R2, D1, KV, Browser Rendering or similar edge or serverless platforms.
Preferred: familiarity with Hono, Next.js, React, Bun, Tauri, Docker and API-first development.
Preferred: experience with PDF extraction, OCR, Docling, PDF.js, browser automation, scraping or content-cleaning pipelines.
Preferred: experience with open-source NLP and ML tools such as Hugging Face Transformers, sentence-transformers, PyTorch, vLLM, Ollama, llama.cpp or similar.
Preferred: familiarity with graph algorithms, entity linking, topic modeling, community detection or visualizing large collections of information.
Preferred: experience with datasets on argumentation, claim-evidence links, debate, research synthesis or scientific literature.
Preferred: experience with safety, trust, provenance, privacy and misuse concerns for AI research tools.
Preferred: contributions to open-source packages, SDKs, APIs, developer tools or technical communities.
Conditions
Work arrangement is flexible. Scope, hours, location and commitment are agreed before you start.
Until the company closes an equity round or secures enough funding, there is no salary, hourly pay or guaranteed cash.
The company values measurable, inspectable and honest AI over features that simply call a large model.
Work is done in the open where practical, so documentation, reproducible work, clear issues and strong pull requests are part of the job.
The company picks tools by what works best: a frontier API when it adds leverage, an open model when control or cost matters, and deterministic code when it is more reliable.
Sources, quotes, provenance and transparent reasoning count as much as fluent answers.
Users keep meaningful control over their data, model choices, memories and research workflows.
Applicants should send a résumé, LinkedIn, GitHub, portfolio or similar; a short note on a production AI, search, agent or open-source project they are proud of; an example of how they would test whether a citation-grounded research agent is improving; and their preferred working arrangement, time zone, availability and pay expectations.
Applicants from nontraditional backgrounds are encouraged to apply if they have built strong systems, shipped useful tools, done meaningful research or contributed to open source.
Benefits
Equity of up to 5 to 10%, with the final amount and vesting set in the agreement.
Target salary of $150,000 a year after successful funding, not guaranteed.
A chance to shape the product, technical direction, fundraising story and early company foundations.