Volume 03 of 05

Projects

Deal support through agentic development.

AI-powered research processes

Specialized agents that produce cited company research at analyst standard. Read onClose
  • Inline hyperlinks in every output, so research is accessible for other parties to go verify. The source trail is the deliverable, not an appendix to it.
  • Results in more thorough and cited research. 49 documents to date. The longest is a 30-page market structure study with 76 footnotes.
  • Company deep dives returning in an afternoon what desk research takes days to assemble.
  • Entity resolution and counterparty mapping, resolving a brand or trade name to the legal entity that holds the contracts, and establishing who a company does business with before a claimed relationship gets underwritten.
  • A full 21-company comparable universe built and normalized in one sitting, so the peer set is internally consistent.

Deck creation

A preset outline that holds tonation and wording constant across board decks, investor materials, CIMs and teasers. Read onClose
  • The outline ensures tone and wording remain the same document to document.
  • Style and formatting are held to one house standard.
  • The writer supplies the content and the argument. The system supplies the rest.
  • Multi-week analyses reach a finished document in days.

Model review

A five-check structured review run against a fixed rubric before a model reaches a decision-maker. Read onClose
  • Assumptions, fundamentals, formula integrity, presentation, and scenario and sensitivity coverage.
  • Findings logged by severity, location, evidence and recommendation, re-checked round over round.
  • Results in better model quality and higher confidence in the numbers reaching the board.

Agentic software development.

Automated testing system

Drives real phones, desktops and browsers, with screenshot and video proof of every step. Read onClose
  • Evidence files are read off disk before a run may report a pass, so results are verifiable after the fact.
  • Every run is fingerprinted.
  • Its own audit surfaced two unwired checks, and a phase was added to close them.

Engineering enforcement layer

Sits between an AI coding agent and anything it can change, so every change clears the same gates. Read onClose
  • Mandatory security gates and live user-flow testing against the running application.
  • A human sign-off step that cannot be bypassed.
  • Runs as the development backbone for every other project here.

Local code knowledge index

An always-on code index queryable by any agent, running entirely on the machine. Read onClose
  • Hybrid keyword and vector retrieval over AST-aware chunks.
  • Cross-language impact analysis.
  • Zero data egress.

ParcelPing. Chain-of-custody verification for e-commerce fulfillment and returns.

Co-Founder · 2024 – 2025

  • Capture package contents at packing, verify against the order, and file a timestamped record either party can rely on.
  • The returns module ran on hardware warehouses already owned and went live in under an hour.
  • Led a team of 11 to first revenue in nine months, recruited largely on equity. Roughly 20 NDAs, two channel partnerships. Conversations reached a major carrier, a global payments network, an e-commerce platform and a third-party logistics provider.
  • A strategic process reached live diligence. Rather than raise a bridge into a contracting market, the company was wound down in October 2025.

Contact

Solomon Dworsky · Miami, Florida

Transaction work in healthcare technology. Applied AI in the investment process.