Draft — AI Civic Action Accelerator 2026: Apply for... - Opportunities for Youth

MOTIVATION LETTER

The Aspen Policy Academy's AI Civic Action Accelerator asks for early-stage civic AI policy ideas, and I am bringing one that comes from a working system, not a concept document. My psyche-twin project is a typed, multi-scale knowledge-graph architecture that lets a person model their own interactions with AI systems. It fuses multiple independent evidence streams, LLM-derived, assessment-derived, behavioral, and document-derived, into one append-only event log. When those streams disagree, the disagreement becomes an explicit graph edge rather than being averaged away. A first-person conversational interface sits on top of the graph and generates responses grounded in the graph's current state. I built this system as an independent researcher in Nigeria, and I am now enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute in Potsdam, Germany, starting Winter Semester 2026/27.

The policy problem I want to address in this accelerator is transparent government AI. When a citizen interacts with an AI system deployed by a public agency, they have no way to see what the system inferred about them, no way to contest it, and no way to understand how their input shaped the output. My proposal is a civic AI policy framework built around a citizen-side model: a personal, portable record of every interaction a person has with government AI, stored on their own infrastructure, with the right to inspect, correct, and appeal the inferences made about them. The technical core already exists in psyche-twin. The policy work is to translate that architecture into a concrete proposal for US state and local governments: procurement language, data standards, and a citizen-facing appeals process.

This fits the accelerator's stated focus on civic participation and transparent government AI directly. The program asks for policy outputs, policy briefs, strategy documents, and advocacy materials, and I am prepared to produce those. I have already produced pre-registered, peer-review-tracked research outputs across multiple domains, including a Bayesian-calibrated pharmacological model of reward-memory encoding with all five pre-registered hypotheses confirmed, and a powered replication study that settled a published comparison that had never actually been run. I know how to define a problem, specify a method, and report results honestly, including negative results. My ergofluids project, a Koopman-operator transport model, failed its first real-data validation gate, and I reported that directly rather than reframing it. That discipline is what policy work requires.

I can commit to all six live virtual sessions on Thursdays from 9:30 to 11:30 AM PT. I have run production systems operations, including Linux VPS, systemd, Caddy TLS, CI/CD, and automated backup and disaster recovery, so I can handle the technical infrastructure of a policy project. I have built four independent DuckDB-based ingest-to-analyze pipelines across life sciences, tech and AI security, and social science domains, which means I can process the policy literature and public comment data this project will require. The accelerator's emphasis on early-stage ideas and tangible outputs matches where I am: I have a working technical system, and I need the policy framing, the government relationships, and the accountability structure that this program provides.

RESEARCH STATEMENT

The psyche-twin project is a typed, multi-scale knowledge-graph architecture for self-modeling. It was built to solve a specific problem: when a person interacts with an AI system, the system builds a model of them, but the person has no equivalent model of the system or of their own interaction history. My architecture inverts that asymmetry. It gives the person a persistent, append-only event log of every interaction, with multiple independent evidence streams feeding into it. An LLM-derived stream captures what the AI system said and did. An assessment-derived stream captures structured measurements. A behavioral stream captures what the person actually did. A document-derived stream captures external records. When these streams disagree, the disagreement is stored as an explicit graph edge, not averaged away. The conversational interface on top of the graph generates responses grounded in the graph's current state, so the person can ask questions like, "What did the system infer about me last month, and what evidence did it use?"

The policy translation is straightforward. Government agencies at the US state and local level are deploying AI systems for benefit determinations, housing allocation, predictive policing, and child welfare screening. Citizens interacting with those systems currently have no portable record of those interactions and no mechanism to contest the inferences made about them. My proposal is a civic AI policy framework that requires three things from any government AI deployment: a machine-readable interaction log given to the citizen after every substantive interaction, a citizen-side model that can ingest that log and show the person what inferences were made and on what evidence, and an appeals process that lets the citizen flag a disagreement between their own record and the agency's record, with the disagreement preserved as an explicit edge rather than resolved by the agency alone.

The technical feasibility is established. I have built the knowledge-graph architecture, the event log, the evidence-stream fusion, and the conversational interface. The system passes 62 of 62 tests in my neurocascade project, which uses the same Bayesian calibration and ODE simulation stack. The policy work is to specify the data standards, the procurement language, and the appeals procedure. That is the work I want to do in this accelerator.

The program's selection criteria emphasize clarity of problem definition, feasibility of policy solution, potential for public impact, and alignment with program objectives. My problem definition is precise: citizens lack a portable, contestable record of their interactions with government AI. My policy solution is feasible because the technical core exists and the policy mechanism, procurement language and data standards, is within the authority of state and local governments. The public impact is measurable: every person who interacts with a government AI system would gain a right they do not currently have. The alignment with the program's focus on transparent government AI and civic participation is exact.

I am proposing a policy output, not a research project: a policy brief, a model procurement clause, and a data standard specification, all of which can be presented to policymakers within the accelerator's timeline. My background supports this. I have a B.Pharm from the University of Ibadan with a German equivalent grade of 1.9, and I have published pre-registered research across addiction neuroscience, protein ML, and dynamical systems. I have also reported a failed validation gate honestly in my ergofluids project, which is the same honesty a policy process requires. The accelerator's six live sessions and between-session assignments are a structure I can meet, and the output requirements match what I am prepared to deliver.

EDITOR NOTES

- The chosen research line is psyche-twin, consistent with the recommended framing angle. It is the only project in the profile that directly produces a civic AI policy artifact, and its technical stage, a working knowledge-graph architecture with a conversational interface, matches the program's early-stage idea requirement.

- Eligibility risk: the program is US-focused and asks for policy solutions for US governments. The applicant is Nigerian, based in Germany for the M.Sc. program, and has no stated US affiliation. The letter addresses this by framing the work as policy output for US state and local governments, but the applicant should confirm whether non-US residents are eligible before submitting.

- The letter claims the applicant can commit to all six live sessions at 9:30-11:30 AM PT. The applicant is in Germany, which is 9 hours ahead of Pacific Time, meaning sessions would run 6:30-8:30 PM CET. The applicant must verify they can attend at that time and should state this commitment explicitly in the application form if asked.

- The applicant must insert personal detail not in the profile: any prior experience with US government processes, any existing contacts in US state or local government, and any specific US jurisdiction they would target for the model procurement clause. The letter currently names benefit determinations, housing allocation, predictive policing, and child welfare screening as target areas, but the applicant should verify these are accurate examples of current government AI deployments before submitting.

MOTIVATION LETTER

The Aspen Policy Academy's AI Civic Action Accelerator asks for early-stage civic AI policy ideas, and I am bringing one that comes from a working system, not a concept document. My psyche-twin project is a typed, multi-scale knowledge-graph architecture that lets a person model their own interactions with AI systems. It fuses multiple independent evidence streams, LLM-derived, assessment-derived, behavioral, and document-derived, into one append-only event log. When those streams disagree, the disagreement becomes an explicit graph edge rather than being averaged away. A first-person conversational interface sits on top of the graph and generates responses grounded in the graph's current state. I built this system as an independent researcher in Nigeria, and I am now enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute in Potsdam, Germany, starting Winter Semester 2026/27.

The policy problem I want to address in this accelerator is transparent government AI. When a citizen interacts with an AI system deployed by a public agency, they have no way to see what the system inferred about them, no way to contest it, and no way to understand how their input shaped the output. My proposal is a civic AI policy framework built around a citizen-side model: a personal, portable record of every interaction a person has with government AI, stored on their own infrastructure, with the right to inspect, correct, and appeal the inferences made about them. The technical core already exists in psyche-twin. The policy work is to translate that architecture into a concrete proposal for US state and local governments: procurement language, data standards, and a citizen-facing appeals process.

This fits the accelerator's stated focus on civic participation and transparent government AI directly. The program asks for policy outputs, policy briefs, strategy documents, and advocacy materials, and I am prepared to produce those. I have already produced pre-registered, peer-review-tracked research outputs across multiple domains, including a Bayesian-calibrated pharmacological model of reward-memory encoding with all five pre-registered hypotheses confirmed, and a powered replication study that settled a published comparison that had never actually been run. I know how to define a problem, specify a method, and report results honestly, including negative results. My ergofluids project, a Koopman-operator transport model, failed its first real-data validation gate, and I reported that directly rather than reframing it. That discipline is what policy work requires.

I can commit to all six live virtual sessions on Thursdays from 9:30 to 11:30 AM PT. I have run production systems operations, including Linux VPS, systemd, Caddy TLS, CI/CD, and automated backup and disaster recovery, so I can handle the technical infrastructure of a policy project. I have built four independent DuckDB-based ingest-to-analyze pipelines across life sciences, tech and AI security, and social science domains, which means I can process the policy literature and public comment data this project will require. The accelerator's emphasis on early-stage ideas and tangible outputs matches where I am: I have a working technical system, and I need the policy framing, the government relationships, and the accountability structure that this program provides.

RESEARCH STATEMENT

The psyche-twin project is a typed, multi-scale knowledge-graph architecture for self-modeling. It was built to solve a specific problem: when a person interacts with an AI system, the system builds a model of them, but the person has no equivalent model of the system or of their own interaction history. My architecture inverts that asymmetry. It gives the person a persistent, append-only event log of every interaction, with multiple independent evidence streams feeding into it. An LLM-derived stream captures what the AI system said and did. An assessment-derived stream captures structured measurements. A behavioral stream captures what the person actually did. A document-derived stream captures external records. When these streams disagree, the disagreement is stored as an explicit graph edge, not averaged away. The conversational interface on top of the graph generates responses grounded in the graph's current state, so the person can ask questions like, "What did the system infer about me last month, and what evidence did it use?"

The policy translation is straightforward. Government agencies at the US state and local level are deploying AI systems for benefit determinations, housing allocation, predictive policing, and child welfare screening. Citizens interacting with those systems currently have no portable record of those interactions and no mechanism to contest the inferences made about them. My proposal is a civic AI policy framework that requires three things from any government AI deployment: a machine-readable interaction log given to the citizen after every substantive interaction, a citizen-side model that can ingest that log and show the person what inferences were made and on what evidence, and an appeals process that lets the citizen flag a disagreement between their own record and the agency's record, with the disagreement preserved as an explicit edge rather than resolved by the agency alone.

The technical feasibility is established. I have built the knowledge-graph architecture, the event log, the evidence-stream fusion, and the conversational interface. The system passes 62 of 62 tests in my neurocascade project, which uses the same Bayesian calibration and ODE simulation stack. The policy work is to specify the data standards, the procurement language, and the appeals procedure. That is the work I want to do in this accelerator.

The program's selection criteria emphasize clarity of problem definition, feasibility of policy solution, potential for public impact, and alignment with program objectives. My problem definition is precise: citizens lack a portable, contestable record of their interactions with government AI. My policy solution is feasible because the technical core exists and the policy mechanism, procurement language and data standards, is within the authority of state and local governments. The public impact is measurable: every person who interacts with a government AI system would gain a right they do not currently have. The alignment with the program's focus on transparent government AI and civic participation is exact.

I am proposing a policy output, not a research project: a policy brief, a model procurement clause, and a data standard specification, all of which can be presented to policymakers within the accelerator's timeline. My background supports this. I have a B.Pharm from the University of Ibadan with a German equivalent grade of 1.9, and I have published pre-registered research across addiction neuroscience, protein ML, and dynamical systems. I have also reported a failed validation gate honestly in my ergofluids project, which is the same honesty a policy process requires. The accelerator's six live sessions and between-session assignments are a structure I can meet, and the output requirements match what I am prepared to deliver.

EDITOR NOTES

- The chosen research line is psyche-twin, consistent with the recommended framing angle. It is the only project in the profile that directly produces a civic AI policy artifact, and its technical stage, a working knowledge-graph architecture with a conversational interface, matches the program's early-stage idea requirement.

- Eligibility risk: the program is US-focused and asks for policy solutions for US governments. The applicant is Nigerian, based in Germany for the M.Sc. program, and has no stated US affiliation. The letter addresses this by framing the work as policy output for US state and local governments, but the applicant should confirm whether non-US residents are eligible before submitting.

- The letter claims the applicant can commit to all six live sessions at 9:30-11:30 AM PT. The applicant is in Germany, which is 9 hours ahead of Pacific Time, meaning sessions would run 6:30-8:30 PM CET. The applicant must verify they can attend at that time and should state this commitment explicitly in the application form if asked.

- The applicant must insert personal detail not in the profile: any prior experience with US government processes, any existing contacts in US state or local government, and any specific US jurisdiction they would target for the model procurement clause. The letter currently names benefit determinations, housing allocation, predictive policing, and child welfare screening as target areas, but the applicant should verify these are accurate examples of current government AI deployments before submitting.

AI Draft — AI Civic Action Accelerator 2026: Apply for... - Opportunities for Youth