Three projects, one pipeline

I built these three projects at different times, for different reasons. The valuation engine came first, then the investment analysis tool, then the publishing pipeline. They all work on their own, but they get more useful when they run in order. Start with a ticker and the last tool hands you an article ready to publish, with no section written by hand.

That connection was not planned from day one. It came from a practical habit I kept repeating. Read a concrete artifact, do the local work, and write something the next tool can understand.

The tools pass files forward instead of sharing state. The valuation engine produces the numbers, Equilyze turns those numbers into analysis, and Pressroom uses the analysis as source material for the final article.

The chain works because the tools stay in their lanes. The valuation engine does not try to write prose. Equilyze does not invent numbers. Pressroom does not rerun the valuation.

• 1
  Equity Valuation EngineDeterministic

  The numbers, deterministic and testable

  Turns one ticker into six valuation models across Bear, Base, and Bull scenarios, plus a composite value and a model-agreement score. No language model touches this layer.

  Project
• 2
  EquilyzeAgentic

  The analysis, agents on top of the numbers

  Wraps the engine, then runs a chain of LangChain agents that add market context, research the news, and combine the material into a recommendation with strengths, risks, and catalysts.

  Project
• 3
  PressroomEditorial

  The story, handled by an editorial pipeline

  Takes the analysis and its artifacts as source material and runs a nine-agent LangGraph pipeline that outlines, drafts, fact-checks, renders charts, and publishes a finished article.

  Project

The Equity Valuation Engine is deliberately the dullest part of the chain. It does math, not judgement. Given a ticker, it normalizes the financial data, decides which valuation models fit the business, and runs the qualifying models across three scenarios. The output is a compact JSON document with every metric, each model’s intrinsic value, the composite, and the agreement score.

Because this layer is deterministic, the same inputs produce the same numbers every time. The agents can argue, summarize, and write, but the valuation stays fixed underneath them.

Equilyze invokes the engine and then interprets the result. A sequence of LangChain agents pulls analyst ratings and momentum, researches recent news, and turns the material into a recommendation. After the first draft, a “curious investor” agent raises follow-up questions, section writers answer them, and a final reviewer reads everything back from disk before signing off.

Equilyze does more than print prose. It writes report.md plus the structured files behind it: valuation_data.json, analysis.json, and news_data.json. Those files are what make the next handoff possible.

Pressroom is a nine-agent LangGraph editorial pipeline. Drop a folder of source material into its input/ directory and it outlines the piece, drafts it, checks it in a bounded reviewer loop, strips AI patterns, designs the frontmatter, renders charts and diagrams deterministically, and writes a finished .mdx file.

For this chain, that folder is exactly what Equilyze produced. Pressroom reads the report and its JSON artifacts and writes the article you can read on this site.

There is no shared library gluing these three together, and no service calls another over the wire. The boundaries are files.

The handoff stays simple. The engine writes JSON, Equilyze adds a Markdown report and more JSON, and Pressroom receives a folder of source material. Each boundary is plain and inspectable, so any stage can be run, tested, or replaced on its own. A person can also open the files and see exactly what passed from one tool to the next.

From a single ticker to a published article, the pipeline follows the same basic path every time.

1. Ticker
2. Value
3. Analyze
4. Artifacts
5. Write
6. Publish

The engine handles the first two steps, Equilyze handles the middle, and Pressroom handles the writing and publication work. Everything runs locally against an Ollama model, so valuation, analysis, and publishing work offline with no cloud API keys.

I could have written this as a single program. Keeping it as three has paid off in ways a monolith would not.

The separate tools still earn their keep outside this pipeline. I can use the engine when I only want a valuation, use Equilyze when I want a research pass, or point Pressroom at an article brief that has nothing to do with finance. The split also makes testing cleaner, because the deterministic core never mixes with the agentic and editorial layers.

They are easier to replace, too. Swap the model behind Equilyze, or point Pressroom at a different brief, and the file contracts still hold. That is the part I like most about the design. The published piece can change tone, structure, and framing without losing the numbers that started it.