Project summary

Problem: Scientific paper production is accelerating, and AI systems are increasingly used to summarize literature, generate reviews, and draft manuscripts. These tools scale our ability to process language, but scientific reasoning is not fundamentally linguistic. The logic of experiments—independent variables, dependent measurements, control conditions, and statistical tests—is typically embedded in dense prose and figures rather than represented explicitly. As the volume of research grows, both humans and AI systems must reconstruct experimental structure manually from text. Scaling language alone is therefore not enough to scale scientific reasoning.Hypothesis Prism is a framework for representing hypothesis-driven experiments as structured, machine-readable objects (https://hypothesisprism.com/). Instead of leaving experimental logic embedded in narrative text, Prism extracts independent variables, dependent measurements, control groups, and statistical tests and organizes them within a formal schema.HP as a solution: This approach creates a necessary structural representation layer for science. By treating experiments as structured objects rather than textual descriptions, Prism makes experimental logic explicit, navigable, and computable. Scientists can explore the core structure of experiments without reconstructing them from dense papers. More importantly, this structured representation enables computational scaling. As experiments are encoded into Prisms, they form a dataset of structured experimental logic grounded in hypothesis testing. Such a dataset could support new forms of scientific navigation, comparison, and AI systems trained directly on experimental structure rather than only on text.

The long-term ambition of Hypothesis Prism is to upgrade the representational medium of science—from narrative descriptions of experiments to explicit structural representations of experimental reasoning.

What are this project's goals? How will you achieve them?

The immediate goal is to refine and scale the existing Hypothesis Prism prototype into a robust pipeline for representing hypothesis-driven research papers as structured experimental objects rather than purely textual descriptions. The framework aims to improve clarity and accessibility of experimental logic, strengthen reproducibility by making methodological structure explicit, and provide a normalized representation of experiments that can eventually support computational analysis and comparison across research.Over the grant period, I will:

1. Expand the structured corpus
   Encode 50–150 hypothesis-driven papers into Prisms across multiple domains (e.g., molecular biology / neuroscience / psychology / clinical or other empirical fields).
2. Systematically resolve edge cases
   Build a documented “edge-case library” and update the encoding pipeline to handle recurring complex structures (e.g., nested variables, conditional designs, multi-stage experiments, atypical control structures).
3. Improve the robustness of the extraction pipeline
   Increase the proportion of papers that can be converted with minimal manual intervention, and track extraction quality against a small “gold” subset of hand-validated encodings.
4. Ship and iterate on a public interface
   Deploy a public-facing Prism explorer that supports browsing encoded papers, navigating variable hierarchies, and inspecting statistically significant relationships.
5. Run a small evaluation with real users
   Recruit 10–20 researchers and test whether Prism improves comprehension of experimental structure (e.g., time-to-answer structured questions about an experiment, or accuracy on structure-identification tasks) compared to reading the paper alone.

This work builds on an existing functional prototype and formal specification. The underlying construction logic is already defined: independent and dependent variables are identified from the paper, organized into hierarchical structures, and connected through experimental design and findings mappings that encode statistically significant relationships. The next phase focuses on refining this pipeline, validating it across a wider range of experimental designs, and building a structured dataset of experiments that demonstrates the feasibility and usefulness of this representation.

How will this funding be used?

Hypothesis Prism is currently being developed as an independent research project. A formal specification, working visualization prototype, and AI-assisted extraction pipeline already exist. The current bottlenecks are development time and the cost of running infrastructure and LLM APIs, especially if the system is opened for broader public testing.

Funding would support:

1. Public testing of the system
   Deploy and maintain a public interface where researchers can explore and test Hypothesis Prism representations.
2. LLM APIs and AI development tools
   API usage and tooling for AI-assisted extraction of experimental structure from research papers (OpenAI, Anthropic, Claude Code).
3. Infrastructure and cloud services
   Compute, storage, database, and hosting needed to support the Prism dataset and public interface.
4. Researcher stipend
   Support one year of full-time work refining the encoding pipeline, resolving edge cases, expanding the structured corpus, and iterating on the public-facing system.

Budget

1. LLM APIs and AI development tools — $8,000
2. Infrastructure and cloud services — $4,000
3. Development tools and operational costs — $2,000
4. Researcher stipend (12 months) — $36,000Total: $50,000

Funding levels (minimum + target)

• Minimum ($10,000): ~3–4 months of full-time work to expand the encoded corpus, harden the pipeline against common edge cases, and ship a stable public demo (v1). ~20–40 encoded papers
• Intermediate ($30,000): ~6–7 months of full-time work to cover more domains/edge cases, improve automation reliability, and run a structured evaluation with early users. ~50–100 encoded papers
• Target ($50,000): ~12 months of full-time work to scale the corpus substantially, iterate on the public interface, and validate the system across a broader range of hypothesis-driven papers. ~100–150+ encoded papers

Who is on your team? What's your track record on similar projects?

Hypothesis Prism is currently a single-researcher project led by me (Aya Samadzelkava).

I trained in mathematics and physics as an undergraduate and MS in neuroscience, where my research focused on network dynamics, gene expression networks, and complex systems approaches to biological questions.

My background spans computational neuroscience, networks, category theory, and topological data analysis, with research experience in biological network modeling, recurrent neural networks, and analysis of large biological datasets in Python.

This background strongly shaped the motivation for Hypothesis Prism. Working across mathematics, physics, and biology repeatedly exposed a gap between the precise structure of experiments and the textual way experiments are communicated in papers. Hypothesis Prism emerged from this intersection of interests in formal structure, complex systems, and empirical research. The project applies tools from mathematics and network thinking to represent the logic of hypothesis-driven experiments as structured objects rather than purely textual descriptions.

I have been developing Hypothesis Prism independently for approximately 1.5 years, including

• formal frameworks
• working visualization prototype
• AI-assisted extraction pipeline for converting research papers into structured experimental representations.

What are the most likely causes and outcomes if this project fails?

The main risks for Hypothesis Prism are that practical or adoption challenges slow its development or limit its usefulness in the near term.

What could go wrong

• Scientific papers are structurally messy.
  Experimental designs vary widely across fields, and converting them into a consistent structural representation requires in-depth understanding of edge cases.
• Extraction may remain partly manual.
  AI-assisted extraction of experimental structure may not become reliable enough for full automation, limiting scalability.
• Public interface costs may be higher than expected.
  Opening the system for public testing could create unpredictable API and infrastructure usage spikes.

Why these outcomes are still valuable

1. A structured dataset of experiments would still be created.
   Even partial success produces a corpus of experiments encoded as structured objects rather than text.
2. The edge-case landscape becomes explicit.
   A documented library of real experimental edge cases and how they map (or fail to map) into structure is valuable for future structural science tooling.
3. The limits of automated extraction become clearer.
   Understanding where AI-assisted extraction fails is useful for future work on scientific knowledge representation.
4. The project would still produce usable artifacts.
   A working prototype, a public demo, and a partially encoded corpus are meaningful outputs even if the system doesn’t reach broad adoption yet.

Conclusion

Scientific knowledge is currently stored primarily as language, even though the underlying logic of experiments is structural. As the volume of research grows and AI systems become more integrated into scientific workflows, relying on text alone will become increasingly limiting. Structural representation of science is not optional—it is necessary for scaling scientific reasoning. Hypothesis Prism is an early attempt to build that structural layer.\

How much money have you raised in the last 12 months, and from where?

As per this date, I have not raised any funds for this project. \