You can sign up instantly with a one-click Google sign-in or set up a traditional account with an email and password.

⚠️ Disclaimer: This is strictly a desktop web experience right now. The high-density chamber maps and timelines need screen real estate, so open this on a laptop or monitor—the mobile layouts and native apps aren’t built yet, but are on the roadmap. Historical congresses are also actively being backfilled as we solidify the graph schema and data ingestion pipelines.

Sign up and start exploring right now:

👉🏽 StringTheory

What’s on the board today

This initial release is focused entirely on the legislative engine — mapping the actual actions of our representatives and the executive branch. It also allows you to follow along as new Bills get sponsored and new Votes occur. Here’s what you can do on the platform starting right now.

The Congress Floor

A fully interactive visual map of the House and Senate chambers. Search any bill or nomination, select a vote, and watch the seats recolor based on who defected, who skipped, or who held the caucus line.

Politicians

Comprehensive profiles of every sitting member and their predecessors back to 1973. Committee assignments, terms of office, an objective ideological placement, and an aisle-crossing engine that isolates every roll call where a politician broke ranks with their party majority.

Bills

Centralized hubs for tracking 119th Congress legislation, featuring full enrolled bill texts and a unified status timeline that interleaves physical chamber actions with logical roll-call votes.

This baseline is just the starting line. Now that the core foundation is stable, I want to take you behind the scenes of the next two core features coming next month: fusing Funding, and building the Agentic Newsroom.

Funding

The first major roadmap milestone is integrating campaign finance into the existing Knowledge Graph. The flow of legislation is only part of the story. In order to run campaigns, get elected and gather resources to execute agendas, you need funding. Corporate interests and influential individuals leverage their wealth to influence voting patterns to enable their interests.

Right now, the data-quality pipeline is in the lab, processing millions of FEC data spanning 46 years of election cycles to present. This includes PAC, party, and super-PAC contributions, alongside independent expenditures.

By mapping this data directly into the property graph, we represent financial contributions in the context of the actions they influenced.

By mapping this data into a network, we move past static spreadsheets to answer three core macro questions

Network Topology (House vs. Senate Profiles)

Because a Senator operates on a statewide or national stage while a Representative defends a hyper-localized district, their underlying funding skeletons are fundamentally different. The graph exposes this contrast instantly:

• The Statewide/National Network: Maps distributed clusters of national elite donors, venture capital pools, and out-of-state holding companies.

• The Regional Network: Tracks highly concentrated local capital—regional manufacturing syndicates, in-state agricultural operators, and localized power brokers.

The Industry-to-Vote Pipeline

Instead of looking at isolated donations, the schema links an individual contributor to an Industry, which bridges directly to a Politician, a Vote, and a specific legislative Topic. This allows us to map systematic sector patterns across whole caucuses, answering the raw operational question: Which industries fund which floor votes?

Committee and Seat Tracking

Lawmakers do not vote in a vacuum; their power is concentrated inside specific committees. By linking the Committee a Politician serves on to the Donor funding their active Race, the architecture maps whether the money flow correlates directly with upcoming regulatory oversight, committee assignments, or changes to bill text before a vote even takes place.

The Newsroom

The second roadmap milestone is more ambitious, but signifies where this product needs to go and the type of functionality that changes how people consume political content.

Search Bar Paralysis

When I showed this beta to a few early testers, the initial feedback exposed a massive structural blind spot. If you are an active investigator who arrives with a specific bill or representative in mind, the platform can directly address immediate specific questions. In its current state, it does very little insight generation, it focuses on exposing the breadth of data in a structured way to help you explore major entities in full detail. But for casual observers who don’t know what they are looking for, raw data exposure becomes intimidating and can cause paralysis.

They don’t have a specific name to type into a search bar. They don’t know which votes matter, which bills are sliding through committee, or where the friction points hide. They arrive with no clear question, meaning they leave with no clear answers.

Agentic SportsCenter

The Newsroom takes trending online narratives and roots them straight to an empirical data layer of real actions. By continuously scanning the web to track what the political ecosystem is actively talking about, the Multi-Agent engine isolates a hot topic—whether it’s a high-stakes vote on Iran war powers or a sprawling corporate tax cut bill—and learns to reason about that public discourse. Instead of adopting the partisan spin, it leverages the StringTheory knowledge layer to surface the hidden connections and signals that intuitively explain what is happening under the hood. The final output isn’t editorial commentary; it’s an interactive, data-driven story packed with verifiable insights, delivered as a low-friction alternative to consuming the conventional news, podcasts or short video content.

The inspiration is OG SportsCenter from 2009-2014, before the takes got hot. Scrolling side panel of stories on the left, a single hero in the middle covering the current story, a live ticker across the bottom keeping you up to date

Case Study: Why did Bill Cassidy change his mind?

I built one exhibit to define the standard. This story unfolded just this past week, and it is live on a preview of the Newsroom on StringTheory right now for you to follow along. This interactive investigative post directly leverages our knowledge graph and was built entirely with Claude Code. This was human-in-the-loop coding, not an autonomous agentic run—it pins down the quality bar the system has to hit on its own, rather than a sample of what it already generates out of the box.

To view the full interactive video playback and trace the live data primitives yourself, you just need to create your free account:

👉🏽 Sign up to interact with this story

This story investigates the 50–47 Senate procedural vote on S.J. Res. 185 (Iran War Powers) where Cassidy voted YAY along with Senate Democrats to remove US Armed Forces in Iran.

The entire exhibit fits in a single viewport with no scrolling allowed to enforce effortless consumption. Alongside the content, the “How it got made” trace exposes the full trail of how the Agents used tools and reasoned about the online narratives and StringTheory data.

The Cassidy exhibit uses four distinct primitives:

• A Three-Paragraph Lede: Covers the vote, the political frame, and what’s next. It names every defector and ties the flip back to three days earlier, when Cassidy lost his Louisiana primary to a President Trump-endorsed opponent. It also flags the three absent Republicans whose return would have killed the resolution by tipping the tally to a 50–50 tie. Links to all mentioned Politicians.

• A Nine-Tick Timeline: Maps Cassidy’s full Iran war-powers voting record this Congress. It exposes that he voted along party lines 8 straight times before suddenly flipping sides on May 19. Every tick is an interactive data visualization linking directly to that specific roll call on the Congress Floor.

• A Two-Column Reading Panel: Contrasts the official narrative against the graph. Column one features the Cassidy Statement—his NBC-sourced quote citing constituent concerns and Pentagon ambiguity. Column two displays What Else Was True—three grounded facts that complete the picture: he lost his primary on May 16, Trump endorsed his opponent back in January, and his term ends if he does not win the Midterms this fall.

• A Loyalty Footnote: Contextualizes the baseline. It tracks Cassidy’s 97.5% party loyalty against the three routine GOP crossers—Collins (73.1%), Paul (74.2%), and Murkowski (78.8%). The graph proves the other three were always statistically likely to cross the aisle. Cassidy was not.

How the system works

Producing a piece like this autonomously breaks into three jobs. The Cassidy walk-through is the easiest way to see what each one has to do.

1. Information Agent — Find the story.

These agents continuously scout the noisy open web, breaking news feeds, and social channels to map what the political ecosystem is actively discussing. They flag current events that are actively being discussed and the narratives around them. The agent's sole job is to consume public discourse, isolate trending stories and hand it to the pipeline as a seed, so the system can strip away the chatter and ground the narrative in valuable data-driven context.

2. Data Analyst Agent — Reason and verify through the graph

Once handed a narrative seed, these agents take over to separate public spin from empirical signal. They break down the story and try to find relevant receipts, walking the graph to isolate the structural intersections behind the news—like bridging Cassidy's roll-call vote directly to his historical voting trajectory (mapping 8 consecutive Nays before a sudden flip on Operation Epic Fury), his exact primary election timeline (losing his seat to a Trump-endorsed opponent three days prior), and his lifetime 97.5% party loyalty model against routine crossers. By checking public statements against these hard ledger rows, they construct an objective, logical way of interpreting the story.

3. Piece Generation Agent — Dynamic presentation that fits the investigation

These agents look at the shape of the insights generated—whether it's a stark voting flip or a hidden financial bridge—and programmatically design intuitive layouts to present the story. These layouts are completely fluid; the Agent functions like a visual editor, determining the most compelling way to structure the piece. It extracts contextual elements from the trending online narrative and seamlessly fuses them with our data-driven backbone. By dynamically assembling the right mix of visual primitives—like timeline blocks, comparison sheets, or video reels—it presents the entire intersection in a highly digestible, high-impact format tailored specifically to that story.

Guardrails

To ensure the autonomous engine never degrades into speculative commentary, three non-negotiable rules govern the pipeline. If a generated piece violates any of these metrics, the system automatically pulls it from the feed:

• Fully Grounded: Every single metric, vote count, and lawmaker name is strictly tagged to direct entities or actions in StringTheory

• Interactive Citations: There are no blind assertions. These pieces will be littered with interactive data visualizations, external links to the story seeds and internal links to StringTheory pages for mentioned entities and actions.

• Zero Adjectives: The system is programmatically banned from using editorialized language like “shocking betrayals” or “controversial bills” unless its direct quotes the characters involved.

There will be future substack articles that fully breakdown how this system is being built through Google Cloud Platform and their suite of Generative AI tools.

Exciting Next Month

V0 for both Funding and the Newsroom ship in June. Sign up for a free account today to try out consuming Politics in a different way. Use it to follow the news or do research. I'll be posting some more video and tutorial content that showcases how I use StringTheory and a lot of the stuff I have already learned from using this tool.

👉🏽 StringTheory

Stay safe during Memorial Day!

— Pravin

But before I hand over the keys, I want to give you a preview of the platform, the current user experience, and the kind of research you can do on day one.

The Invisible Engine: Why a Graph?

In my first post, I mentioned StringTheory is built like a digital corkboard. That wasn’t just a metaphor for the UI—it’s a literal description of the underlying architecture.

StringTheory is powered by a Knowledge Graph, specifically built on Spanner Graph. I chose this architecture because it gives us a unique superpower: it combines the rock-solid reliability and massive scalability of a world-class relational database (Spanner DB) with the native ability to map and traverse complex networks. I also have never used Google Cloud Platform before, and I am using this project to learn GCP, so that made this an easy choice.

Most political tracking sites rely entirely on traditional databases, where data lives in rigid, isolated tables (e.g., a table for Politicians, a table for Bills). That structure is great for simple questions like, “List all the Senators from Ohio.”, where you are doing simple CRUD operations on a relational database through an API. But we want to ask complex investigative questions: “Show me every politician who co-sponsored these three specific defense bills, and then crossed the aisle to vote against their own party on this related amendment.” In a purely traditional database, querying those multi-layered connections requires massive, clunky equations that effectively hit a wall. But by leveraging Spanner Graph, our data acts as a continuous network. Everything is a Node (a politician, a bill) permanently wired together by Edges (voted for, sponsored by). We aren’t just cross-referencing tables; we are telling the engine to walk the relationships.

Structuring the underlying data as a Graph from day one unlocks the long-term value of StringTheory:

• Complex Investigations: Natively trace influence and voting blocs through multiple degrees of separation (”multi-hop” reasoning) in milliseconds.

• Predictive Analytics: Train Graph Neural Networks (GNNs) on the “shape” of a politician’s network to predict how they will vote or detect hidden coalitions before they go public.

• Deep Research AI: Instead of an LLM hallucinating a narrative based on op-eds, we can ground generative AI directly in our graph. It won’t guess; it will explore the connections intuitively and be able to reason in more complex ways and hand you the receipts.

For this initial Beta, we are laying down this foundation. Before we can track the money or build advanced analytics, we need to map the actual actions. So, V0 of StringTheory is purely the legislative engine.

Here is a walkthrough of the initial platform and how you can use this tool to answer some questions today.

1. Politicians

Most investigations center around our elected officials and their actions. The main power we have as citizens is to analyze our representatives’ actions and hold them to account when actions don’t align with promises or with the priorities of their constituents.

The first step is figuring out who your representatives are or discovering persons of interest you want to explore.

The Politicians page provides a list view to be able to search and sort through current and historical members of Congress, Presidents and Vice Presidents.

In List View, you can sort on party loyalty, bipartisan sponsorship rates, bills introduced. There are easy filters for chamber, party and even who is on the upcoming ballots for the Mid-terms.

If you don’t know who you are looking for, Map View can help you identify persons of interest by state.

Once you have found your legislator, you can click on their profile page, which gives you the immediate legislative footprint of a single elected official.

You aren’t just reading a curated biography. You can instantly see their objective ideological placement based purely on their roll-call votes, their recent floor activity, and their closest legislative collaborators. It strips away the campaign rhetoric and focuses entirely on their measurable track record in the chamber.

In modern politics, the most telling votes are the ones that break ranks, because in a polarized environment most votes are along party lines. Instead of guessing who the swing votes are, StringTheory isolates and highlights when a politician voted with the opposing majority instead of their own caucus. Republicans currently have more Congress members than the Democrats, so for contested votes to pass, you need defectors, and those are often the interesting policy propositions.

2. Congress Floor

Speaking of voting, the Congress Floor is where the action happens.

We mapped the Senate and House as they actually sit. If you go back to previous congresses, you can see how party control in congress has fluctuated under different administrations. You can also pick any bill, resolution, or nomination, and watch the chamber color-code exactly how the room voted. It turns a boring roll-call list into an immediate, visual heatmap, which will be highly sharable for receipts in content and research.

Instantly, the partisan divides become visually obvious. You can filter the floor by specific topics—like “International Affairs,” “Defense,” or “Economics”—and also search on the title and content of your target legislation. Oftentimes, the same piece of legislation will have multiple votes as it passes through each chamber, and we visually showcase this through timelines on both the Congress Floor and the individual Bills page.

3. Bills

Instead of relying on editorialized summaries from the news, StringTheory allows you to search and track bills directly through the system.

When you click into a specific piece of legislation, it acts as a central hub. You can see its entire lifecycle timeline. You instantly see the primary sponsors, co-sponsors, and the tagged topic areas for that bill.

Scroll down, and the graph isolates exactly who voted against their party’s majority. And you can read the full text of the bill itself without ever leaving the page. On the roadmap is building advanced Natural Language Processing (NLP) metadata enrichments, so we can pull out more nuanced concepts, topics, keywords from the full raw bill text, versus someone’s opinion on passed legislation.

4. The Executive Branch

The legislative machine doesn’t stop at Congress, and StringTheory also tracks Executive Branch officials, namely the President and Vice President.

You can track Executive Orders and bills signed into law, allowing you to see how unilateral actions fit into the broader political landscape. You can also see on VP pages when the VP has voted as a tie-breaker in the Senate.

But the Executive and Legislative branches collide most visibly during the nominations process.

StringTheory maps these confirmation battles directly on the Congress Floor view. When the administration appoints a new cabinet secretary or federal judge, you can watch exactly how the chamber whipped the votes, revealing which politicians align with the White House.

What’s Next & The Beta Launch

We will be rapidly adding more complex features and analytics to this graph (campaign finance data is highest priority), but this legislative foundation is the necessary first step.

I’m dropping the beta link right here on Substack this weekend for anyone to sign up and try it out. You will be able to search for your own representatives, track the bills you care about, and pull the threads yourself.

See you this weekend.

— Pravin

What used to take a team of six engineers six months now takes one person a week. Sometimes a day.

With the massive leaps in agentic coding and frontier LLMs, the traditional moats of software development have evaporated. You no longer need millions in funding to compensate a massive headcount of data scientists, ML engineers, and frontend devs just to get an MVP out the door. Today, the only real limit is your product sense.

I’ve spent my career in data science and applied ML, shipping products across industries that have nothing to do with each other—from the Kansas City Royals front office, to national defense, to sports prediction markets, and currently leading public sector Generative AI research at Unstructured. Working across sports, media, and government taught me that your effectiveness isn’t bound by domain expertise. It’s about translation. The real superpower of applied ML is taking a messy business problem and translating it into a concrete technical solution. That skill transcends the industry. If you can do that, you can build anywhere—and AI just put that ability into overdrive.

Introducing PQ Labs

That observation is why I’m launching PQ Labs.

PQ Labs is the umbrella brand for the projects I’m building across these industries. The goal is simple: do real research, build novel SOTA applications of ML and AI, and ship them as real products. Pure research often misses the strongest signal that something is actually valuable—someone pulling out their wallet to pay for it. I’m not here to write papers. I’m here to ship products with real business models, and I’m using this Substack to document the entire process in public.

Content creation has always felt daunting to me. But in a world where the algorithm runs everything and we all live in echo chambers, doing the work in public and promoting the real exchange of ideas matters more than ever.

Enter StringTheory

For the last few weeks, I’ve been heads-down building the first product coming out of PQ Labs: StringTheory.

Every crime drama or investigative thriller has that scene—the detective standing in front of a map, photos pinned up, red strings crisscrossing to connect people to places, dates, and motives. By creating a network of the players, the bank accounts, and the locations, a follow-the-money investigation reveals the hidden patterns that explain the whole operation.

That is exactly what this product is. It’s the digital version of that map, built to track the United States political system.

Every politician, bill, donor, and vote is a pin. The strings between them are the connections—who funded whom, who sponsored what, who voted how, who sits on which committee. Pull a thread, and the pattern reveals itself.

Why this, why now

I chose this as the first project because the whole point of building solo right now is velocity. Moving fast is the advantage you have at any startup, and what’s the use of AI tools that let you build a product in a week if you just end up spending the next six months trapped in a slow, miserable enterprise sales cycle? I wanted to build something I could put directly into the public’s hands on day one and see what happens.

More importantly, it’s a tool I desperately wanted to build for myself. I need this to exist.

The state of the political landscape in 2026 is the worst I’ve seen in my lifetime. Good-faith conversation has been crippled by collapsing trust in legacy outlets and a new independent media space dominated by influencers with their own agendas and self-interests. Between “fake news” rhetoric and algorithmic confirmation bias, the public’s understanding of what our elected officials actually do is at an all-time low, including my own.

The thesis is simple: Public trust is at an all-time low. Today, any editorialized content—whether it’s a TikTok, a podcast, a newspaper column, or directly from a politican themselves—is instantly dismissed as propaganda. That reflex kills any chance of a productive discussion about what's actually happening in this country.

Between a genuine passion for the underlying data and my own frustration with this current toxic climate, StringTheory was the perfect starting line. It pushes back against the noise. It is grounded entirely in measurable actions pulled from primary sources the government is legally obligated to publish. I can’t fix journalism or the algorithm, nor do I know how, but I can build a tool that helps find signal in the noise through data-driven insights that are not opinions, but puzzle pieces that help you reason objectively about your own views.

Under the hood, this system is a Knowledge Graph, which is structured in a way that maintains the vital relationships between entities, legislation and funding required to connect the dots.

Which donors funded the sponsor of this bill?

Which committee chairs hold positions that correlate with their campaign funding?

How have a senator’s views evolved over time, told through their voting history and PAC donations?

Have you ever wondered any of these things? I have. And right now there is no easy way to understand these patterns and get clear answers on what has happened in the past or predicting what will happen in the future.

Who is the target user

• The Knowledge-Hungry Voter: You want to understand your state’s delegation and see the source material for yourself before deciding what you think.

• The Investigative Journalist: You need to trace connections across systems that weren’t designed to talk to each other, to conduct a data-driven investigation and discover why things happen the way that they do.

• The Social Media Comment Warrior: You want objective, shareable receipts to drop into an argument. Real primary-source evidence that holds up under scrutiny, not my facts vs. your facts.

The Tenets

You can’t build in a space this noisy without a North Star. These four tenets are the absolute foundation of StringTheory. They are a public commitment to staying on mission, and every single feature shipped will answer to them:

• Actions, not rhetoric. Votes cast, checks cashed, bills signed. What someone does is the signal; what they say is mostly noise.

• Federally mandated sources. We rely entirely on data the government is legally obligated to publish. When that data is messy, delayed, or incomplete, we will be radically candid about the gaps. But the rule remains: no receipts, no post.

• The full tally, not the half. Every cosponsor, every donor, every defector. Not just the slice that fits a narrative.

• Facts, not verdicts. Quantitative evidence to build your own case on. Not judge, jury, or executioner.

Where this goes

What launches next Friday is the MVP—useful on day one. But the long-term vision is to build complex analytics on top of this foundation, including vote prediction with graph neural networks, natural-language Q&A over the graph, and coalition detection to surface the hidden voting blocs that actually pass or kill legislation.

As the Knowledge Graph foundation becomes more solid, these won’t be just ideas on a whiteboard. We’ll tackle them one by one—taking a real use case, translating it into a technical model, and showing exactly how it gets built.

The Rollout

• Monday, April 27: Product walkthrough with screenshots.

• Friday, May 1: Launch day. StringTheory goes public, I’ll drop the domain, and a tutorial video on some of the key features.

• Ongoing: Fridays are for talking product, tech, and tools. Mondays are for showcasing the tool and its features by investigating historical events and analyzing what’s going on in US politics today.

If you’re building something ambitious by yourself, the Friday posts are for you.

If you’re open to consuming political content and analysis in a completely different way than editorialized content, the Monday posts are for you.

Thanks for being here on day one.

— Pravin