Pico Probe: Multi-Model Research Pipeline for AI-Aided Math

What the company appears to be: Pico Probe is a self-reported research infrastructure tool for AI-assisted mathematics. The author describes it as a system that orchestrates AI math research through multi-model, verified pipelines, building an investigation as a graph of mathematical knowledge with provenance, replay, and unexplored paths.

What changed: The project evolved from a simple multi-model UI into what the author describes as "a research infrastructure" — a system that represents mathematical investigations as persistent graphs rather than sequences of prompts or text exchanges.

Single most important open question: Is Pico Probe a working prototype or a conceptual framework? The description states it was built using Codex and ChatGPT, but there is no evidence of actual deployment, usage, or validation in real research workflows. The author makes claims about functionality (e.g., "Negative Knowledge", structured communication) that are not demonstrated.

Commercial due-diligence read: This appears to be a single-person project submitted for a hackathon. It lacks any evidence of traction, revenue, customers, or adoption. The description is self-reported and unverified — it contains no data on product-market fit, user feedback, or commercial viability. The author's own write-up suggests the tool is conceptual and experimental in nature.

The description states that Pico Probe is research software for AI-assisted mathematics. It is described as a system that:

• Builds an investigation as a graph of mathematical knowledge, complete with provenance, verification, and reusable failed approaches.
• Treats an investigation as a computational object composed of typed mathematical artifacts, explicit dependencies, execution history, provenance, and verification records.
• Represents research as a persistent graph rather than prompt chaining or text exchange.
• Uses a Research Kernel to compile visual pipelines into executable dependency graphs.
• Implements a research protocol with globally unique identifiers, timestamps, dependency information, provenance, and structured mathematical content.
• Supports capability-based routing of tasks (e.g., symbolic simplification to SymPy or Mathematica).
• Provides a Research Graph that stores mathematical artifacts, semantic relationships, and execution history among various nodes (LLMs, Python code, Lean, SymPy, etc.).
• Includes a plugin system for AI models, theorem provers, symbolic engines, simulations, and custom research tools.
• Enables replay, verification, and extension of investigations over time.

Inference: The author describes Pico Probe as a system that builds an investigation into a structured graph, not just a tool to orchestrate AI interactions. It is positioned as a persistent knowledge layer for mathematical research.

The description states that Pico Probe is not another multi-agent framework, but rather a system that builds an investigation as a graph of mathematical knowledge and prioritizes the preservation of scientific reasoning.

It claims to be a research infrastructure, not just a UI or pipeline orchestrator, and positions itself as a persistent layer above models, theorem provers, and symbolic computation systems.

The author also states that Pico Probe is inspired by Git’s approach to code history but applies it to AI-assisted mathematical reasoning. It aims to make the evolution of mathematical knowledge persistent, rather than treating research as prompt engineering.

Inference: The positioning has evolved from a simple multi-model UI into a conceptual infrastructure for managing and preserving mathematical knowledge. However, this evolution is not demonstrated in the description — it remains self-reported.

The description states that Pico Probe is intended for researchers working in AI-assisted mathematics, particularly those dealing with problems involving geometric probability, polylogarithms, or other complex mathematical domains where coordination of models and tools is difficult.

It targets users who want to:

• Preserve reasoning history.
• Reuse failed approaches.
• Inspect evidence behind claims without rerunning the entire project.
• Replay research processes chronologically.
• Extend investigations months later without reconstructing previous work.

Inference: The target customer is likely a mathematician, researcher, or advanced AI user working in formal mathematics or symbolic computation. However, no specific customer segments or personas are identified.

Not evidenced.

The description does not contain any information about pricing, monetization, or business model. There is no mention of customers, revenue, or commercial use cases beyond the author’s own research.

The description states that Pico Probe was built using:

• Codex and ChatGPT 5.6 Sol
• Python, JavaScript, CSS, VSCode
• A Research Kernel that compiles visual pipelines into executable dependency graphs
• A research protocol with typed messages, globally unique IDs, timestamps, provenance, and structured content
• Execution envelopes that provide upstream/downstream dependencies, graph state, and execution history
• A plugin system for AI models, theorem provers, symbolic engines, simulations, and custom tools
• A Research Graph storing mathematical artifacts, semantic relationships, and execution history
• Capability-based routing of tasks to specific tools (e.g., SymPy for symbolic simplification)
• Negative Knowledge — failed reasoning paths are preserved in the graph

It also mentions:

• API key storage using Fernet encryption
• Secure database storage in SQLite, with Docker containerization
• Browser login tokens stored in localStorage

Inference: The system is built on a combination of modern AI tooling (Codex, ChatGPT), Python-based backend, and structured communication protocols. However, there is no evidence of production deployment or scalability.

Not evidenced.

There is no mention of:

• Customers
• Revenue
• Adoption
• Usage metrics
• Product-market fit
• Iteration history or feedback loops
• Deployment in real-world research settings

The description states that the project was submitted to a hackathon, and that it evolved from a simple multi-model UI into a research infrastructure. However, no evidence of actual use or impact is provided.

The author mentions several existing tools that Pico Probe builds on or complements:

• LangChain
• Model Context Protocol (MCP)
• LeanDojo
• SymPy
• Jupyter

These are described as frameworks that coordinate execution, but not as systems that provide a unified representation of the evolving research process.

Inference: Pico Probe positions itself as a complementary layer to these tools, rather than a replacement. It is intended to add persistence and structure to existing AI-assisted math workflows.

1. No evidence of real-world use or adoption: The project was submitted to a hackathon and lacks any demonstration of actual usage.
2. Self-reported functionality without validation: Claims about "Negative Knowledge", structured communication, and replay are not demonstrated.
3. Single-person development: The team size is listed as 1, suggesting limited resources for product development or scaling.
4. No commercialization strategy: No pricing, monetization, or customer acquisition plans are described.
5. Unclear maturity level: The author describes it as a "research infrastructure" but does not provide evidence of its current state (prototype, alpha, beta).
6. Dependency on AI hallucinations and prompt chaining: While the system aims to reduce these issues, it still relies on AI models for core functions.

1. What is the current maturity level of Pico Probe? Is it a prototype, alpha, or something more?
2. Has it been tested in real research environments? If so, what were the outcomes?
3. How does it handle model disagreement or conflicting outputs from different models?
4. Can you demonstrate how the Research Graph is built and replayed?
5. What are the limitations of the current implementation?
6. Are there any plans to integrate with existing math research platforms or tools (e.g., Jupyter, Lean)?
7. How does it ensure reproducibility across different environments or model versions?
8. What is the roadmap for product development and commercialization?