SEQ-01: SEQUEN.io

Brief description of the organization

SEQUEN is a mobile-first skin health intelligence platform operated by 12754941 Canada Inc., a Toronto-incorporated company. The product sits as the data infrastructure layer between consumer skincare and clinical dermatology, capturing longitudinal skin imagery and correlating observable changes against hormonal cycles, sleep, treatments, ingredient exposure, and environmental data. Outputs use strictly observational language and never offer diagnostic claims.

The platform launches on iOS in July 2026 with two revenue lines: a B2C subscription for individual users and a B2B clinic integration tier for dermatology and aesthetic medicine practices. The technical stack includes computer vision (MediaPipe 478-landmark detection across 13 anatomical zones), an LLM-backed observational scoring engine, a Fitzpatrick-stratified personalization system, and a Canadian-region Supabase backend built for PIPEDA compliance.

SEQUEN's R&D roadmap spans four phases: V1 launch, V1.5 severity scoring and condition detection, V2 proprietary skin-specific foundation model, and V3 controlled-capture hardware paired with multi-modal biosensor integration. The company is preparing for a pre-seed raise in Q3 2026, with active engagement in IRAP, SR&ED, and Mitacs Accelerate pathways. Built for anyone who cares about their skin, SEQUEN is designed for scientific defensibility from day one.

Problem area

The problem space: Skin is the largest organ and the most accessible biomarker of internal physiology, yet consumer skincare and clinical dermatology operate in disconnected data silos. Consumers track outcomes anecdotally. Clinicians see snapshots at six-month intervals. No continuous, structured longitudinal layer exists to capture what is actually happening to skin over time and correlate it against the inputs that shape it: hormonal cycles, sleep, products, treatments, ingredients, and environmental conditions. Two layered failures sit underneath this silo problem, and SEQUEN is built to bridge both.

Failure one: the diagnostic equity gap. Current dermatology infrastructure was built for, and on, light skin. The Fitzpatrick Skin Type scale, still the dominant classification system in clinical and research practice, was developed in 1975 to predict UV sensitivity in white patients undergoing PUVA therapy. The original scale was Types I to IV, all white skin. Types V and VI were appended later, collapsing the full diversity of brown and Black skin into two oversimplified categories. The scale is also self-reported, based on burning and tanning behaviour rather than measured pigmentation, which makes it subjective and inconsistent across users. The downstream consequences are well-documented in the literature. Dermatology training image datasets are heavily skewed toward Types I to III. Commercial dermatology AI tools have measurably lower accuracy on darker skin tones. Conditions like post-inflammatory hyperpigmentation, melasma, keloid scarring, hidradenitis suppurativa, and acral melanoma present differently on skin of colour and are routinely missed or misclassified. Standard clinical photography protocols (lighting, white balance, exposure) are calibrated for lighter skin and lose texture and chromatic detail in darker skin. Skin cancer is diagnosed later in patients of colour, and outcomes are worse. The harm is real and measurable.

Failure two: the temporal and contextual gap. Even when diagnosis is accurate, current dermatology operates on six-month in-clinic snapshots, subjective clinician-eyeball severity scoring, and almost no correlation against the lifestyle, hormonal, or environmental inputs that drive change. Skin is dynamic. A snapshot misses everything that happens between visits.
How SEQUEN bridges. Five design principles run through the platform. First, baseline-as-self: each user is compared against their own longitudinal baseline rather than against a biased population norm. Second, per-zone analysis across thirteen anatomical regions rather than a single global skin classification, since condition response varies by site. Third, observational language only, never diagnostic, so the system reports what it sees ("texture appears smoother in zone 3") rather than claiming knowledge it doesn't have. Fourth, inclusive dataset construction from day one rather than retrofitting diversity after launch, with Fitzpatrick used as a transitional input while the platform builds toward more objective measures (Individual Typology Angle, Monk Skin Tone Scale). Fifth, multi-signal correlation that reduces reliance on visual data alone, where the historical bias lives. The R&D roadmap spans four phases, each with distinct engineering questions a student team could anchor on.

V1: Reliable longitudinal capture from a consumer phone. The first technical problem is capture quality. Phone cameras vary across white balance, exposure, lens characteristics, and user behaviour. Lighting, angle, distance, and skin orientation drift session to session. The current build uses MediaPipe Face Landmarker (478 3D landmarks, 13 anatomical zones) for spatial normalization and Anthropic Claude Vision for observational analysis. Open questions: per-zone signal extraction under inconsistent illumination, colour constancy across device generations, capture-quality scoring to gate user feedback, and Fitzpatrick-stratified preprocessing that explicitly compensates for training bias toward lighter skin tones. The longer-horizon V1 question is replacing self-reported Fitzpatrick input with objective skin-tone measurement (Individual Typology Angle, Monk Skin Tone Scale) extracted from the capture itself.

V1.5: Temporal change detection and severity scoring. Once capture is reliable, the next problem is meaningful change over time. SEQUEN tracks six condition charts (texture, pigmentation, erythema, hydration, elasticity, and a composite). The scoring engine is currently v2 with seven personalization layers. A team could focus on multi-zone temporal differencing across longitudinal sessions, severity stratification defensible to a dermatology advisory board with skin-of-colour expertise, condition detection that holds across the full skin-tone range (post-inflammatory hyperpigmentation, melasma, keloid response, and other presentations underserved by current models), and correlation strength between observable skin signals and structured user inputs like hormonal cycle phase, sleep, and ingredient exposure.

V2: Skin-specific foundation model. Post-raise, the moat is a proprietary LLM trained on SEQUEN's longitudinal corpus rather than relying on general-purpose vision models. The engineering problem is dataset curation with deliberate skin-tone balance, training methodology for a small-domain foundation model, and benchmarking against general-purpose baselines on dermatologically relevant tasks. Teams with ML grounding could explore self-supervised pretraining strategies, multi-modal fusion (image plus structured logs plus biometrics), or evaluation frameworks that hold across the full skin-tone spectrum with explicit Fitzpatrick I to VI and Monk 1 to 10 stratification.

V3: Accessible biomarker capture and the skin-hormone correlation engine. The long horizon is closing the loop between observable skin changes and the internal physiological signals that drive them. Hormonal cycles are the most direct driver of cyclical skin behaviour, yet hormone tracking remains gated behind blood draws, expensive lab panels, or single-use ovulation strips. Recent advances in saliva-based hormone monitoring (lateral flow immunoassays, salivary ferning pattern analysis, colorimetric cartridges read by smartphone camera) make continuous multi-hormone tracking technically viable for at-home use without dedicated reader hardware. The capstone problem is the accessible biomarker capture layer that feeds SEQUEN's correlation engine: a disposable saliva collection format, a smartphone-readable assay or ferning pattern, a computer vision pipeline that returns hormone level estimates from a phone camera image, and integration with SEQUEN's longitudinal skin dataset. A parallel V3 thread is multi-modal fusion with wearable biosensor data already in market (sleep, heart-rate variability, skin temperature) to model the skin-brain-stress axis. Recent work on tri-modal brain encoders (Meta TRIBE v2, March 2026) is directly relevant to the fusion architecture.

A team could anchor on any phase. Strongest fits are biomedical and chemical engineering with smartphone computer vision (V3 accessible saliva capture), computer vision and ML with explicit fairness-across-skin-tone evaluation (V1 and V1.5), data engineering with privacy architecture (cross-phase), or clinical validation methodology with skin-of-colour dermatology advisor benchmarking (V1.5). The shared thread across every phase is the bridge SEQUEN is trying to build: from snapshot to longitudinal, from biased to inclusive, from anecdotal to structured, and from visual-only to multi-modal.

Main objectives

The engagement has one primary objective: deliver a scoped engineering project that bridges one of SEQUEN's four named gaps and produces a working prototype, validated methodology, and integration path into SEQUEN's V1.5 production launch in Q4 2026. A team will anchor on one of four tracks, selected during the project framing phase based on team composition and discipline strengths.

Track 1 objectives: Fairness-aware computer vision and longitudinal change detection (V1 and V1.5).

• Build an objective skin-tone classifier from facial capture that outputs Individual Typology Angle and Monk Skin Tone position, validated against ground-truth measurement across at least 200 paired samples spanning all Monk tiers.
• Quantify per-zone signal reliability across illumination conditions for SEQUEN's thirteen anatomical zones and deliver a capture-quality gating algorithm that flags unusable sessions.
• Develop and benchmark a multi-zone temporal differencing model for longitudinal change detection across six condition signals (texture, pigmentation, erythema, hydration, elasticity, composite).
• Evaluate model performance with explicit Fitzpatrick I to VI and Monk 1 to 10 stratification, treating fairness across the full skin-tone range as a primary acceptance criterion rather than a post-hoc check.
• Deliver a working integration with SEQUEN's React Native and Supabase stack, deployable into V1.5 production.

Track 2 objectives: Severity scoring and clinical validation (V1.5).

• Design a severity scoring protocol for SEQUEN's six condition charts defensible to a dermatology advisory board with skin-of-colour expertise.
• Build a validation framework comparing SEQUEN scoring outputs against dermatologist ground-truth annotations on at least 100 longitudinal case series across the full Fitzpatrick range.
• Quantify inter-rater agreement, intra-rater stability, and AI-versus-clinician concordance using standard statistical methods (Cohen's kappa, ICC).
• Identify condition categories underserved by current AI baselines on darker skin (post-inflammatory hyperpigmentation, melasma, keloid response, and others) and produce a remediation roadmap.
• Deliver a whitepaper-quality methodology document and protocol package ready for IRB submission.

Track 3 objectives: Privacy-preserving longitudinal data architecture (cross-phase).

Design and prototype a PIPEDA-compliant Canadian-region data architecture supporting longitudinal storage, mobile ingest, and B2B clinic export.

• Implement differential privacy on aggregated population queries while preserving per-user fidelity for individual longitudinal analysis.
• Build a consent management layer with granular user control over export, retention, and clinical sharing.
• Define and stress-test the API surface for B2B clinic integration with documented latency and throughput characteristics.
• Deliver a working migration path from SEQUEN's current Supabase configuration to the new architecture.

Track 4 objectives: Accessible biomarker capture and integration (V3).

• Design a saliva collection and assay format compatible with smartphone-camera readout, with a disposable consumable cost target under $5 per test.
• Develop a computer vision pipeline that returns calibrated hormone level estimates (cortisol first, with extension path to progesterone or estradiol) from a phone-camera image of the assay.
• Quantify accuracy and reproducibility against laboratory ELISA reference standards across at least 50 paired samples.
• Build an integration prototype that ingests hormone readings into SEQUEN's longitudinal user record and correlates against existing cycle-phase and skin-imagery data.
• Produce a patent landscape analysis covering Eli Health, Salignostics, Ovul.ai, and related prior art, with a defensible IP path for SEQUEN's coupling layer.

Cross-cutting acceptance criteria applicable to any track:

• User-facing outputs use strictly observational, non-diagnostic language.
• Evaluation frameworks include explicit fairness-across-skin-tone stratification as a primary acceptance criterion, not a retrofitted check.
• Code deliverables integrate cleanly with SEQUEN's React Native, TypeScript, and Supabase stack.
• Scientific deliverables are of publication or whitepaper quality, suitable for use in SEQUEN's pre-seed and Series A fundraising.

Scope of work

Engagement model. Eight months of work across four phases, with one four-person interdisciplinary team anchoring on one of the four tracks named in the objectives section. Track selection happens within the first six weeks based on team composition, faculty advisor expertise, and a project framing session with SEQUEN. The work then proceeds through discovery, design, build, and validation, culminating in a deployable prototype and validated methodology handed off to SEQUEN's V1.5 production roadmap.

Phase 1: Discovery and track scoping (weeks 1 to 8). The team onboards with SEQUEN's codebase, design system, and existing R&D documentation. Mutual NDA and IP assignment agreements are executed during onboarding. The team conducts a focused literature review for the selected track and stakeholder interviews with SEQUEN's founder, R&D lead, clinical advisor (Mount Sinai Hospital, Obstetrics and Gynecology), and a sample of NDA-cleared beta testers. The phase ends with a signed track scope document defining the prototype concept, validation protocol, acceptance criteria, sample size targets, and a risk register.

Phase 2: Design and methodology (weeks 9 to 16). The team produces detailed technical design (architecture diagrams, algorithm specifications, schemas, or hardware and assay design as appropriate to the track) and an experimental or validation protocol. For Track 2, IRB submission materials are drafted. For Track 4, a prior art and patent landscape review is delivered, with input from SEQUEN's external legal counsel (Osler Hoskin & Harcourt). An initial v0.1 prototype is built and demonstrated at the midterm review. Phase deliverable: design document plus working v0.1.

Phase 3: Build and integration (weeks 17 to 26). The team builds the full v1 prototype against the design specification. For computer vision and ML tracks, this includes model training, hyperparameter tuning, and evaluation on the assembled dataset. For data architecture, this includes schema implementation, API surface, and stress testing. For biomarker capture, this includes assay fabrication, smartphone-camera readout calibration, and ELISA reference comparison. Integration with SEQUEN's React Native, TypeScript, and Supabase stack is delivered where applicable. Phase deliverable: working v1 prototype plus integration documentation.

Phase 4: Validation, documentation, and handoff (weeks 27 to 32). The team executes the full validation protocol against the acceptance criteria defined in Phase 1, with fairness-across-skin-tone stratification as a primary gate. Statistical analysis is performed against pre-registered methods. A whitepaper-quality final report is produced, suitable for inclusion in SEQUEN's pre-seed and Series A fundraising materials. The team presents at the Waterloo Capstone Symposium. Code, methodology, data, and IP are transferred to SEQUEN per the executed assignment.

In scope: Track-specific objectives delivered to defined acceptance criteria. Integration with SEQUEN's existing technical stack where applicable. Validation against anonymized beta cohort data under NDA. Engagement with SEQUEN's clinical advisor on protocol design. Engagement with SEQUEN's external legal counsel on patent landscape and IP questions. Standard Waterloo Capstone deliverables (symposium presentation, final report, peer review).

Out of scope: SEQUEN's production launch infrastructure for V1 in Q3 2026. App Store-facing user experience and visual design (handled by SEQUEN internally). B2B clinic sales or business development. Fundraising activities. Long-term commercial deployment, manufacturing setup, or regulatory submission (handled by SEQUEN post-capstone). Direct interaction with end users outside the validated NDA-cleared beta cohort.

Resources provided by SEQUEN: Access to existing codebase via NDA-bound GitHub repository. Mutual NDA and IP assignment templates executed during onboarding. Access to anonymized beta cohort data and existing SEQUEN scoring engine outputs. Introduction to the clinical advisor at Mount Sinai Hospital. Weekly one-hour partner check-ins with the founder and biweekly technical reviews with the R&D lead. Reasonable compute budget for ML training (Tracks 1 and 2). Capped consumables and fabrication budget for biomarker capture (Track 4), negotiated at project framing. Introduction to external legal counsel (Osler) for patent landscape and IP questions.

Resources required from Waterloo: A four-person interdisciplinary team with disciplines matched to the selected track. Faculty advisor with relevant domain expertise. Standard Waterloo Capstone time commitment over eight months. Compliance with SEQUEN's NDA and IP assignment terms. Access to standard Waterloo lab, prototyping, and computing facilities.

Deliverables

• Report
• New protocols/processes

Team meeting frequency

Weekly.

Skills and training required

The project draws on a mix of technical, scientific, and professional skills, with track-specific emphasis. Some are expected coming in, others are developed through the engagement.

Baseline skills useful across all tracks. Strong fundamentals in research and analytical methods, experimental design, statistical analysis, technical writing, and presentation. Comfort working in interdisciplinary teams, managing scope when requirements shift, and operating under NDA and IP-assignment frameworks. Project management with milestone tracking against defined acceptance criteria. Stakeholder communication across founder, clinical advisor, technical lead, and external legal counsel.

Track 1: Fairness-aware computer vision and longitudinal change detection. Computer vision fundamentals including image processing, colour science, and segmentation. Machine learning methodology covering training, evaluation, hyperparameter tuning, and dataset curation. Algorithmic fairness analysis with subgroup stratification. Working knowledge of Python, PyTorch or TensorFlow, MediaPipe or equivalent landmark frameworks. Mobile integration in React Native and TypeScript. Familiarity with skin-tone classification systems (Fitzpatrick, Individual Typology Angle, Monk Skin Tone Scale).

Track 2: Severity scoring and clinical validation. Clinical research methodology including IRB submission processes. Statistical analysis for inter-rater reliability (Cohen's kappa, intraclass correlation coefficient) and concordance studies. Protocol design and documentation to publication standard. Familiarity with dermatology clinical practice and the literature on skin-of-colour presentations (post-inflammatory hyperpigmentation, melasma, keloid response, hidradenitis suppurativa). Whitepaper authoring suitable for fundraising and academic publication.

Track 3: Privacy-preserving longitudinal data architecture. Backend and cloud architecture on Postgres-based systems (Supabase preferred). Differential privacy techniques and anonymization methodology. API design with documented latency and throughput characteristics. Privacy law literacy covering PIPEDA, GDPR, and HIPAA principles. Database schema design and migration. Consent management system patterns.

Track 4: Accessible biomarker capture and integration. Biomedical engineering with focus on lateral flow immunoassay or saliva ferning pattern analysis. Chemistry of antibody-antigen interactions and colorimetric reactions. Smartphone computer vision for colour detection, calibration, and exposure normalization. Mechanical design for a disposable cartridge or collection format with manufacturability considerations. Prior art and patent landscape analysis methodology. ELISA reference methodology for laboratory benchmarking.

Skills developed through the engagement (any track): Direct, contemporaneous experience inside a venture-stage health tech company, with exposure to founder, technical R&D lead, clinical advisor, and external legal counsel. Health tech regulatory awareness, particularly the distinction between observational and diagnostic framing. Equity-centred engineering practice, with fairness-across-skin-tone evaluation treated as a primary acceptance criterion rather than a retrofitted check. LLM-assisted product development through hands-on integration with Anthropic Claude APIs. Fundraising-quality deliverable production, including whitepaper and methodology documentation suitable for investor due diligence. Cross-functional collaboration with NDA-cleared beta testers and clinical partners.

Resources required 

Resource needs vary by track. Some are provided by SEQUEN, others draw on Waterloo's standard capstone infrastructure, and a few line items will need negotiation at project framing depending on the selected track.
Computing and cloud (all tracks). Standard development environments, version control (GitHub, provided), and project management tooling (Notion, provided). Tracks 1 and 2 additionally need GPU compute access for model training and evaluation, sourced from Waterloo's existing ML infrastructure or a SEQUEN-funded cloud allowance (AWS, GCP, or equivalent). Anthropic Claude API access is provided by SEQUEN throughout the engagement.

Specialized software

• Track 1: PyTorch or TensorFlow, OpenCV, MediaPipe SDK, iOS Simulator with Xcode, React Native development environment, image annotation tools (Label Studio, CVAT, or similar).
• Track 2: Statistical software for inter-rater reliability and concordance analysis (R, SPSS, or Python statsmodels), REDCap or similar platform for IRB-aligned data capture, annotation tooling for dermatologist case review.
• Track 3: Supabase CLI, Postgres tooling, API stress-testing frameworks (k6, Locust, or similar), differential privacy libraries (Google DP, OpenDP, or similar).
• Track 4: OpenCV and image processing libraries for colorimetric readout, CAD software for cartridge design (SolidWorks, Fusion 360, or similar).

Specialized tools and equipment

• Track 1: iOS devices for capture testing (multiple iPhone generations preferred), controlled lighting setups, colour reference standards (X-Rite ColorChecker), and Monk Skin Tone reference cards.
• Track 2: Standard computing environment; the primary need is dermatologist-annotated longitudinal data and statistical tooling.
• Track 3: Standard development and database tooling; no specialized lab equipment required.
• Track 4: Wet lab or biomedical engineering lab access for assay fabrication, microfluidic prototyping equipment, ELISA reference instrument or third-party laboratory service for hormone benchmarking, spectrophotometer for validation, saliva collection consumables, 3D printing access for cartridge prototyping, and standard biosafety equipment.

Datasets and data access. SEQUEN provides access to its anonymized beta cohort data under NDA, including existing scoring engine v2 outputs and longitudinal session metadata. Track 1 may benefit from supplementary public datasets including Fitzpatrick17k, the Diverse Dermatology Images (DDI) dataset, and SD-198 for pretraining or fairness evaluation. Track 2 will need access to dermatologist-annotated longitudinal case series, sourced through SEQUEN's clinical advisor relationship at Mount Sinai Hospital. Track 4 will need ELISA reference samples for hormone validation, sourced through a third-party laboratory service.

Human and advisory resources. Weekly one-hour partner check-ins with SEQUEN's founder, biweekly technical reviews with the R&D lead, and scheduled consultations with the clinical advisor at Mount Sinai Hospital (all provided by SEQUEN). Introduction to SEQUEN's external legal counsel (Osler Hoskin & Harcourt) for patent landscape and IP questions on Track 4 (provided by SEQUEN). A Waterloo faculty advisor with discipline-relevant expertise (Waterloo).

Budget items requiring negotiation at project framing. Cloud compute allowance for ML training (Tracks 1 and 2), SEQUEN-funded up to an agreed cap. Consumables for biomarker capture covering saliva collection kits, assay materials, ELISA reference panels, and cartridge fabrication (Track 4), SEQUEN-funded up to an agreed cap. Hardware purchases including skin-tone reference standards (Track 1), SEQUEN-funded. All budget items are capped and agreed in writing before Phase 2 begins. Waterloo retains responsibility for standard student facility access, faculty advisor allocation, and lab space within existing institutional arrangements.

NDA or a commercialization agreement for this project?

Yes