foundation
Manifest
A document intelligence engine. FastAPI + PostgreSQL + pgvector. Ingests content, embeds it, runs vector search, and returns sourced answers with citations. Not a chatbot.
Read the architecture →Intelligence infrastructure
I turn fragmented information into structured, searchable, actionable intelligence.
I design and build systems that ingest scattered documents, web data, and raw text, then make them searchable, sourced, and ready to act on. Manifest is the foundation. The verticals are built on top.
The thesis
Information is everywhere. Intelligence is rare.
Most useful knowledge is scattered across documents, pages, and inboxes in a form you cannot query. I build the layer that closes that gap, end to end.
Input
Fragmented
PDFs, URLs, raw text, reports, rosters. Real signal, trapped in formats nothing can search.
Core
Structured
Chunked, embedded, and indexed into a vector store. The corpus becomes addressable.
Output
Actionable
Sourced answers with citations. Searchable, defensible, and ready to decide on.
Playbook
Athletics recruiting intelligence. In development.
Blackletter
A second vertical. Exploratory.
Aletheia
reasoningThe reasoning layer. Multi-source synthesis and analysis on top of the foundation.
Manifest
Ingest, embed, search, and cite. The infrastructure every vertical runs on.
Hover or tap a layer to see what it provides, and what the layer above inherits.
Architecture, not a portfolio
One foundation. Many verticals.
This is not a pile of unrelated projects. It is one system, built in layers. Manifest handles acquisition and retrieval. Aletheia turns retrieval into reasoning. Verticals like Playbook apply that intelligence to a specific market.
Build the foundation once. Every vertical inherits ingestion, embeddings, search, and sourcing for free, and adds only what its domain needs.
Verticals are referenced here as direction. Public demos ship when they are ready.
Inside the foundation
How Manifest thinks
01
Acquire
Ingest documents, URLs, and raw text. Parse, clean, and capture metadata.
02
Structure
Chunk content and embed it into a pgvector store for semantic addressing.
03
Remember
Track sources and entities so context and provenance survive across queries.
04
Reason
Run vector search and synthesis to answer questions, not just match keywords.
05
Deliver
Return sourced, cited output through an interface a stranger can navigate.
Evidence
The thesis, proven in code
Shipped tools you can open today, plus the foundation they pointed toward.
shipped
Triager+
A focused ML tool that classifies help-desk tickets by category and priority. FastAPI backend, hand-built static UI. Proof that a model only matters once it ships.
Open the tool →
shipped
ShapeSound
A small domain-specific language that turns symbolic input into visuals and audio. An exercise in parsers, compilers, and symbolic systems, running entirely in the browser.
Open the tool →Verticals in development: Aletheia (reasoning), Playbook (athletics recruiting intelligence), Blackletter. No public links yet. See the full architecture →
Writing
Notes from the build
Jun 2026
From fragmented to structured: the thesis
Why I build intelligence infrastructure, and what separates an engine from a chatbot.
Jun 2026
Inside Manifest: ingest, embed, search, cite
The architecture of the foundation, and the decisions behind each layer.
Jul 2025
Triager+: from concept to execution
Shipping a minimal, memory-aware ML tool end to end, from model to UI.
Building in the open.
I build systems that turn information into an advantage.
If you have information problems costing you time, clarity, or a competitive edge, the foundation is being built for exactly that.