Precise AI tools for precise work
Amphibian builds AI for advisory firms, boutiques, and practice groups whose work is too document-specific, too regulated, and too judgment-heavy for broad AI tools. We partner with the domain experts who know the work.
Trust signals
Our thesis
Most of AI's value hasn't been built yet.
The application of AI to actual work is one of the largest opportunities in software, and most of it is still ahead. The model labs themselves, and the AI-native services firms forming around them, are pointed at the largest enterprises in the world. Vertical AI products are claiming whole industries — legal, healthcare, real estate, financial services. Both layers are real and important.
Both also leave most of the actual work uncovered: the specialty practices and firms inside every industry, where the documents, regulations, and judgment calls are specific enough that broad tools can't do them justice, and where the economics of going deep look different from the economics of going broad.
That's where Amphibian builds. We pick a practice where the work is too specific for general tools and too narrow for the industry-wide players, partner with someone who has spent their career inside it, and build the AI that fits the work. Then we do it again, in another practice.
Case study
Spherient advisors rely on detailed Excel workbooks to compare insurance renewal options for their clients. The source material arrives as long carrier PDFs, with rates, enrollment, and plan details spread across different documents and formats.
Amphibian is building precise AI tooling that reads those documents, structures the data, routes it through advisor review, and produces the final workbook in Spherient’s existing format.
Complex documents become a reviewed workbook.
1. Source documents
Advisors upload the carrier materials they already receive from insurers.
-
Carrier renewal materials
Rates and plan changes across long PDFs
-
Plan summaries and proposals
Benefit details in inconsistent formats
-
Enrollment records
Who is enrolled, in which plan, at what tier
2. Structured review
AI structures the key information, then keeps an advisor in control of the final data.
3. Final workbook
The output is not a summary or a chat response. It is a pixel-perfect Excel workbook that matches Spherient’s existing template, so it drops directly into the team’s workflow without reformatting or adjustment.
Pixel-perfect template
Matches Spherient’s Excel format
No workflow change
Fits the process advisors already use
Ready-to-use .xlsx
Formulas and formatting included
workbook.xlsx The result is not a generic AI assistant.
It is a reliable document-to-workbook system for a high-stakes expert workflow — with humans reviewing the data before it becomes client-facing.
How Amphibian works
AI built with the people who know the work.
Most specialty AI is sold to an industry. Amphibian is built with an industry.
Each specialty product at Amphibian has a domain partner — a senior practitioner or firm who shapes what gets built, leads distribution into their network, and shares the upside. We bring the technology and the product engineering. They bring the deep knowledge of the work, the trust of the buyers, and the distribution.
The result is a product grounded in real operating knowledge from day one — built around the workflow, introduced through trusted relationships, and refined with the people who use it.
Domain partner brings
- Workflow expertise Knows the work at operator depth
- Buyer trust Has credibility inside the niche
- Aligned upside Shares revenue or equity
Amphibian brings
- AI infrastructure Filesystem agents and specialty AI patterns
- Product engineering Design, build, ship, iterate
- Platform connectors Integration surface from prior builds
Together we ship
A product with workflow fit, distribution, and defensibility.
-
Workflow fit
Used by the partner’s team
-
Distribution
Introduced through trusted relationships
-
Moat
Specific to the niche and workflow
Technology
AI for work that has to stay connected.
In specialty firms, the work does not live in one document. It moves across PDFs, spreadsheets, emails, messages, prior decisions, and final deliverables. One source explains another. One row depends on one clause. One output has to trace back to the evidence behind it.
Amphibian is built for that kind of work. We keep the documents, data, decisions, and outputs connected, so AI can follow the workflow instead of treating each source like a standalone page of text.
Read how the platform helps us shipSee how the same work is represented.
Breaks the work into snippets it can search, then has to infer how those snippets relate.
Vector RAG pipeline
Rich data sources become fragmented
PDF clause
renewal.pdf
Word document
benefits-summary.docx
Spreadsheet row
enrollment.xlsx
Slack thread
Salesforce record
account.json
Before you play
A typical AI pipeline starts with rich sources, then retrieves nearby text fragments.
- 01 Extract text from every source.
- 02 Split documents, rows, threads, and records into fragments.
- 03 Retrieve similar language, then rebuild object relationships later.
PDF clause
renewal.pdf
Word document
benefits-summary.docx
Spreadsheet row
enrollment.xlsx
Slack thread
Salesforce record
account.json
RAG ingestion
Extract text
Isolate fragments
Embed
For domain experts
If you know a workflow this should exist for — we should talk.
Amphibian works with senior practitioners — managing partners, founding advisors, practice leaders — who see the document and judgment work in their field that broad AI is getting wrong, and want to build the AI that gets it right. We bring the technology and engineering. You bring the workflow expertise and the network. We share the upside.
What we look for
- 01 A specific workflow in a specialty practice where broad AI fails.
- 02 A trusted network in that specialty — not just adjacent industries.
- 03 Willingness to share risk and upside, not just collect a fee.
Bring the workflow. Bring the trust. We’ll bring the platform to turn it into a product.
Tell us about your workflow