MOTIVATION LETTER The Sacramento AI-Native Startup Accelerator asks for everyday problem solvers, and the problem I bring is specific: mental health support that cannot scale because it depends on one-on-one human attention. I am Eniola Ayodele Olutogun, a pharmacist and computational researcher from Nigeria, and I have spent the last two years building the technical foundation for a solution. My project, psyche-twin, is a typed, multi-scale knowledge-graph architecture for self-modeling. 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, generating responses grounded in the graph's current state. The architecture is the product, not a wrapper around a chatbot. I built it because my clinical work as a pharmacist at Ramset Pharmacy and my research in addiction neuroscience at the University of Ibadan showed me the same gap: patients need continuous, structured self-reflection, but the tools they have are either static questionnaires or ungrounded conversational agents that forget everything between sessions. psyche-twin does not forget. Every interaction is appended to an event log, and every response is generated from the current state of that log. The accelerator's criteria emphasize clarity of the problem, feasibility of the solution, and community impact. The problem is clear: the World Health Organization estimates a global shortage of mental health workers, and Nigeria, where I am from, has fewer than 300 psychiatrists for over 200 million people. The feasibility is demonstrated by my existing implementation. I have built four independent DuckDB-based ingest-to-analyze pipelines across life-sciences and social-science domains. I self-host local LLM serving with llama.cpp and on-demand model swapping. I run production systems on Linux VPS with systemd, Caddy TLS, CI/CD, and automated backup and disaster-recovery. The knowledge-graph core of psyche-twin is already typed and functional. What I need from the accelerator is the structure to turn a working architecture into a tested product with a defined user base in Sacramento. I am currently enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute and University of Potsdam, starting Winter Semester 2026/27. I am willing to relocate to Sacramento for the program duration. The accelerator's no-AI-experience-required framing matches my situation in reverse: I have deep AI experience, but I need the accelerator's network to understand the mental health deployment context, regulatory pathways, and community partnerships that a technical researcher does not automatically possess. I am applying to learn those pieces, and I bring a working system, a clinical background, and a research record that includes three sole-authored preprints under peer review and a co-authored paper under review at Alcohol (Elsevier). RESEARCH STATEMENT I am selecting psyche-twin as the core venture for this application because it is the only line of my work that is both AI-native and directly deployable as a product within the accelerator's timeframe. My other research lines, the CCT model for addiction, TOPOLOGIX for drug-resistance prediction, and neurocascade for brain-circuit simulation, are scientific instruments. They produce papers and validated models, but they do not produce a user-facing service. psyche-twin does. The technical thesis of psyche-twin is that a person's mental state is a multi-scale structure, not a single scalar or a single embedding. At the lowest level, there are discrete events: a message sent, a mood logged, a sleep record, a therapy note. At the next level, there are patterns across those events: a user who always logs anxiety after late-night social media use. At the highest level, there is a self-model: a typed graph that represents the user's own beliefs about their patterns, which may or may not match the measured patterns. The core design decision is that disagreement between streams is preserved as an explicit edge, not averaged away. If a user's self-report says "I am sleeping well" but their behavioral stream shows 4.5 hours of sleep per night, the graph contains both facts and an edge labeled "discrepancy." The conversational interface can then surface that discrepancy directly: "You reported sleeping well, but your sleep log shows otherwise. Which one is accurate?" This design is grounded in my research background. My CCT model in addiction neuroscience is a tripartite pharmacological framework with three coupled axes, dopaminergic reward prediction error, NMDAR-dependent long-term potentiation, and affective contrast, modeled as a system of ODEs and calibrated with Bayesian MCMC using PyMC DEMetropolisZ across 14 free parameters. All five pre-registered hypotheses were confirmed. That work taught me how to build coupled dynamical models where multiple signals interact. psyche-twin applies the same philosophy to personal data: multiple streams, explicit coupling, and no premature reduction to a single number. The current validation status is honest: the architecture is built, the event log is append-only and typed, and the conversational interface generates responses grounded in the graph state. What has not been done is a real-user deployment with behavioral data fits. The circuit-layer parameters in my neurocascade engine are explicitly labeled illustrative pending real behavioral-data fits, and I apply the same standard to psyche-twin. I am not claiming product-market fit or revenue. I am claiming a working system that is ready for a structured pilot with a defined user group. For the Sacramento accelerator, the pilot path is concrete. The first deployment target is a community mental health clinic or a university counseling center in the Sacramento area. The pilot would enroll 20 to 30 participants over eight weeks. Each participant uses the psyche-twin interface daily. The outcome measures are engagement rate, self-reported utility scores, and qualitative analysis of whether the discrepancy edges produce useful insights. The technical infrastructure already supports this: Supabase/Postgres for the event log, DuckDB for analysis, and local LLM serving for privacy-sensitive inference. No cloud dependency, no third-party data sharing. The accelerator's emphasis on community impact maps directly to this pilot. Sacramento has a growing population and a documented need for scalable mental health support. A conversational AI companion that is grounded in a persistent, typed self-model is a testable intervention. If the pilot shows engagement and utility, the next step is a broader deployment. If it does not, the append-only log gives us the data to understand exactly why. SHORT-ANSWER ESSAY: PROBLEM AND SOLUTION The problem is that mental health support is bottlenecked by human attention. A therapist can see a limited number of patients. A crisis line can handle a limited number of calls. Digital tools exist, but most are either static content libraries or ungrounded chatbots that produce fluent text with no memory of prior sessions. Neither approach builds a model of the user over time. psyche-twin solves this by making the model the product. Every interaction, whether it is a chat message, a mood rating, a sleep log entry, or an imported document, is appended to a typed event log. Multiple evidence streams are fused into a knowledge graph. Discrepancies between streams are preserved as explicit edges. The conversational interface reads the graph and generates responses grounded in its current state. The result is a system that remembers, that notices contradictions, and that can ask the user to resolve them. My qualifications for building this are direct. I am a licensed pharmacist with a B.Pharm from the University of Ibadan. I have built and validated Bayesian-calibrated dynamical models in addiction neuroscience. I have built four independent data pipelines using DuckDB for ingest-to-analyze workflows. I self-host local LLM serving with llama.cpp. I run production systems with CI/CD and automated backups. I know how to build the system, and I know the clinical context it must serve. SHORT-ANSWER ESSAY: COMMITMENT AND COACHABILITY The accelerator asks for commitment to the program and coachability. My commitment is demonstrated by my track record of finishing what I start. The CCT model went from a literature screen of 1,847 records to a fully calibrated Bayesian model with all five pre-registered hypotheses confirmed, and it is now under review at three peer-reviewed journals. The TOPOLOGIX project went from a falsified hypothesis about interface topology to a working sequence-based predictor with AUROC 0.804 on the Platinum benchmark. When my ergofluids project failed its first real-data validation gate, I reported the failure directly rather than reframing it. That is the behavior the accelerator should expect from me: I do the work, I report the results honestly, and I pivot when the data says to pivot. Coachability means I will take direction on the parts I do not know. I know the technical architecture. I do not know the Sacramento mental health provider landscape, the California regulatory environment for digital health tools, or the fundraising expectations of a US-based accelerator. Those are exactly the areas where I will follow the program's guidance. I have already demonstrated the ability to learn new domains: I moved from pharmacy to computational modeling to software engineering, and I am now enrolled in the M.Sc. Digital Health program at the Hasso Plattner Institute. I am applying to this accelerator because it explicitly welcomes applicants without AI experience, which tells me the program is structured to teach what participants need. I need the program's help with deployment, not with the core technology. CHECKLIST - [ ] Complete the online application form at https://www.humanbulb.org/accelerator-application - [ ] Verify the application deadline on the programme website - [ ] Confirm Sacramento residency or relocation willingness is stated clearly in the application - [ ] Attach or paste the MOTIVATION LETTER as the primary statement - [ ] Attach or paste the RESEARCH STATEMENT as the technical description - [ ] Attach or paste the SHORT-ANSWER ESSAY on problem and solution - [ ] Attach or paste the SHORT-ANSWER ESSAY on commitment and coachability - [ ] Prepare a one-page technical architecture diagram of psyche-twin (event log, knowledge graph, conversational interface) in PDF format - [ ] Prepare a list of 3 references: Kent Berridge (Michigan), Samuel Gershman (Harvard), Nathaniel Daw (Princeton), with contact details - [ ] Verify ORCID (0009-0001-9272-6735) and GitHub (github.com/AmunRaPtah) links are live and accessible - [ ] Confirm enrollment status at Hasso Plattner Institute / University of Potsdam for Winter Semester 2026/27 with a document or acceptance letter - [ ] Draft a 2-minute pitch video if the programme requests one; confirm from the website EDITOR NOTES - Eligibility risk: The programme is described as open to individuals or teams based in or willing to relocate to the Sacramento area. The applicant is currently in Nigeria and enrolled in a German M.Sc. program starting Winter 2026/27. Confirm whether the accelerator's timeline overlaps with the M.Sc. schedule and whether relocation to Sacramento is feasible for the full program duration. This is the single largest risk in the application and must be resolved before submission. - Fact verification needed: The applicant's employment history lists "National Product Manager, Synthcare (Mar 2026-present)" and "Clinical Pharmacist, Ramset Pharmacy (Jan-Mar 2026)". These dates are in the future relative to a typical 2025 application cycle. Verify the current date and adjust employment dates accordingly. Do not submit with future-dated employment. - Fact verification needed: The profile states the applicant is "Enrolled, M.Sc. Digital Health, Hasso Plattner Institute / University of Potsdam, Germany (Winter Semester 2026/27)". Confirm whether enrollment is confirmed or pending. If pending, state it as "accepted" or "applied" accurately in the application. - Gap to fill: The applicant must insert a specific Sacramento connection or rationale. The current draft states willingness to relocate but does not name any Sacramento-specific community partner, clinic, or university. Before submission, the applicant should identify at least one concrete Sacramento organization (e.g., a community mental health clinic, UC Davis Health, or a local nonprofit) to name as a pilot partner or outreach target. - Gap to fill: The application form may ask for a team composition. The applicant is a solo founder. If the programme prefers teams, the applicant should either state a plan to recruit a co-founder during the program or identify a collaborator who can serve as an advisor. The current draft does not address team structure. - The psyche-twin project is described as a venture, but it has no real-user validation yet. The draft correctly avoids claiming product-market fit or revenue. Keep it that way in the final submission. Do not let enthusiasm in the pitch video or verbal presentation overstate the validation status.