No matching projects. Try another phrase or select All work.
Voice & dealership workflowsImplementation & deployment routing
AUTOVOX
Keep the conversation connected to the work.
A voice-and-workflow product for dealerships, connecting real-time conversation handling with lead context and follow-up.
Read the product decision
The problem
Calls and follow-up lose context when the conversation, customer record, and next action sit in separate systems.
The decision
Treat interruption, timing, and handoff as product behavior. Keep provider readiness and cost distinct from what the interface promises.
What exists
A TypeScript call-session implementation with transcription and speech adapters, media events, interruption handling, and operational follow-up boundaries.
Evidence & boundary
Current source and an archived build corroborate the implementation. Deployment routing is documented; live provider operation, answered-call counts, conversion lift, and revenue are not claimed.
Shared product platformCustomer workspace implemented
DESK
One workspace. Different jobs to be done.
A shared customer workspace for vertical products, bringing onboarding, readiness, team access, billing, integrations, and product-specific activity into a consistent experience.
Read the product decision
The problem
A growing product family needs a coherent customer experience without forcing every industry into the same workflow.
The decision
Separate the shared account experience from domain-specific product logic. Show what still needs setup before presenting a service as ready.
What exists
Signup and login routes, product adapters, workspace settings, team controls, billing states, integration surfaces, and account-scoped data-plan checks.
Evidence & boundary
Current implementation separates tenancy and billing from product setup and activity. This does not establish successful live payments, active customer counts, or validated predictive models.
Research & decision supportImplemented · pilot documented
AI Hour / TraceQuorum
Research you can trace back to the passage.
A workspace that turns authorized source material into cited research and decision reports, preserving assumptions, uncertainty, and the evidence behind each claim.
Bring evidenceAuthorized sources
→Inspect the answerClaims & citations
→Make a decisionReport & assumptions
Read the product decision
The problem
A polished answer is difficult to trust when its sources, gaps, and assumptions disappear. Analysts need to inspect the reasoning and carry it into a usable deliverable.
The decision
Keep the simpler research route as the default until a more complex one earns its extra latency. Return an explicit insufficient-evidence state when the source material cannot support an answer.
What exists
Passage-level citations, bounded retrieval and repair, interrupted-run recovery, source deletion, decision packs, and reproducible report and presentation exports. The implementation uses Python, FastAPI, React, and LangGraph.
A tradeoff, measured.In a saved nine-case synthetic evaluation, the bounded route met all case checks versus eight for the baseline. Its p95 latency was 19.907 ms versus 5.788 ms, exceeding the configured 2× limit. The baseline stayed the default.
Local software evaluation, July 31, 2026. These results do not measure real-world accuracy or customer outcomes.
Current boundary
A single-server pilot is documented. Current uptime, customer adoption, and live external-model execution are separate verification questions.
Agent workflowsImplemented · export-only
Enterns Content Studio
A proposal should survive a useful argument.
A content planning system where specialists develop a cited proposal, critics challenge it, and deterministic checks decide whether it is ready for human review.
Read the product decision
The problem
Specialized contributors can produce inconsistent plans. Sources get lost, metrics get mixed together, and a draft can look ready before anyone checks its claims.
The decision
Give the disagreement a structure. Preserve proposals, challenges, rebuttals, and dissent before drafting. Separate creative approval from permission to publish.
What exists
A LangGraph workflow with 12 specialist roles, five critic roles, cited debate, source checks, and a human review pause. Observed-performance inputs retain their platform and measurement window so unlike metrics are not compared.
Evidence & boundary
Source and tests cover parallel branches, joins, review and resume, and provenance checks. These are software roles, not team size. The default runs offline, and even approved output remains an export-only package. No publishing or growth outcome is implied.
Learning & career developmentApplication implemented
Enterns
Connect assessment to a next step.
A learning-to-work product with distinct experiences for learners, mentors, and organizations, connecting assessment, development, and participation.
Read the product decision
The problem
An assessment alone does not give a learner a path forward. Learners, mentors, and organizations need different tools around a shared development journey.
The decision
Design those experiences separately and connect their workflows. Keep availability and outcome promises tied to verified supply rather than the breadth of the interface.
What exists
Student assessment and pathway surfaces, mentoring, competency testing, and enterprise workflows in a React application. The adaptive-assessment integration authenticates users and reports unavailable dependencies explicitly.
Evidence & boundary
Application routes and integration source demonstrate the built scope. They do not establish current internship availability, assessment validity, placement results, or the completeness of every production flow.
Service-business intakeIntake & handoff implemented
ShopDesk
Make a missed call actionable.
A service-business intake workflow that turns conversations into structured job packets and clear owner handoffs.
Read the product decision
The problem
An unanswered call or incomplete request gives a busy shop little to act on.
The decision
Capture the details the operator needs, then report each delivery channel separately. One failed notification should not conceal the outcome of the other.
What exists
Structured job-packet builders, configuration and storage boundaries, and separate SMS and email notification results with independent error handling.
Evidence & boundary
Source and test files support the workflow. Trade-average ticket amounts in the original product are estimates, not evidence of recovered revenue. No call or notification volume is claimed.
Insurance intakeDomain workflow implemented
AgencyDesk
Capture the request. Preserve the professional handoff.
An insurance intake product that structures requests and makes the boundary between administrative assistance and professional judgment explicit.
Read the product decision
The problem
Agencies need usable intake without an assistant implying that it can decide coverage or make other professional determinations.
The decision
Build the handoff into the workflow. Distinguish rules that can check text before use from transcript review that happens after a provider has spoken.
What exists
Structured intake for first notice of loss, certificate requests, and billing questions, alongside text rules and transcript auditing.
Evidence & boundary
Implementation evidence supports intake and review. Post-call detection is not guaranteed prevention of unsafe speech, and this project is not presented as regulatory approval or licensed advice.
Evidence-grounded disclosureDrafting & review pipeline implemented
Disclosure Sentinel
A draft needs more than fluent language.
A disclosure workflow connecting source-grounded drafting, deterministic checks, provenance, and a professional review workbench.
Read the product decision
The problem
Technical disclosure loses accountability when the source behind a sentence disappears during drafting.
The decision
Gate the draft before persisting its sentence-level record. Keep the source, the checks, and the professional review attached to the work.
What exists
A LangGraph sequence for context, drafting, checks, provenance binding, and audit, supported by ingestion modules and reviewer-facing contracts.
Evidence & boundary
Source and test files cover generation, provenance, audit, and review boundaries. Software rules are not regulatory approval, professional signoff, or proof that a filing is complete.
Evidence & inspection toolsVerification workflow implemented
CertProof
Make a check inspectable.
A certificate-review tool combining structured intake, deterministic anomaly checks, and a traceable report for receiving inspection.
Read the product decision
The problem
A pass or fail result offers little help when the reviewer cannot inspect the supporting records or see which rule raised a concern.
The decision
Keep each rule result, its audit event, and the chain of supporting records together. Route anomalies for review.
What exists
Structured certificate intake, tenant-scoped records, deterministic checkers, per-rule audit entries, and provenance-chain output in a related Sentinel package.
Evidence & boundary
The implementation produces inspectable software checks. It does not establish certificate authenticity, airworthiness, aviation-authority approval, or completed customer inspections.
Commerce & mobilePre-commerce · mobile foundation
NORI & OAK
A clear experience before an open checkout.
A staged headless-commerce and mobile foundation that separates a consumer brand concept from the operational readiness to sell and serve customers.
Read the product decision
The problem
A convincing storefront can accidentally promise stock, subscriptions, or services that are not yet ready.
The decision
Make release boundaries part of the product. Gate catalog, cart, and checkout independently, with required dependencies and clear customer-facing status.
What exists
A Next.js brand experience, typed Shopify integration boundaries, claim policies, and a React Native information foundation with shared contracts.
Evidence & boundary
Source and test files cover feature policy, commerce transport, and mobile boundaries. The mobile foundation is unsigned and non-production. This case makes no claim of working subscriptions, released mobile commerce, or AI capability.
Product education & commerceStorefront implemented
ORENDA Mobility
Make the offer as clear as the product.
A product education and commerce experience with explicit availability checks, consent-aware measurement, and clear failure states across inquiry and checkout.
Read the product decision
The problem
Product presentation, inventory, and the ability to accept an order can move at different speeds.
The decision
Use actual product imagery and let verified availability govern commerce. Keep an inquiry distinct from an accepted order and preserve customer input when a form fails.
What exists
A Next.js storefront, movement education, inquiry and lead routes, and a cart handler that checks activation, product selection, and availability before creating checkout.
Evidence & boundary
The source shows those checks. Current inventory, fulfilled orders, conversion lift, and revenue are not claimed here.
Internal systemsArchitecture & implementation
SABI AI
Make automated work inspectable.
An internal system for routing work, retrieving departmental context, checking supporting evidence, and recording decisions.
Read the product decision
The problem
Automation becomes hard to operate when nobody can tell why a task ran, which evidence it used, or what happened afterward.
The decision
Use explicit work types, provider routing, evidence checks, and a traceable event history. Keep deterministic tasks in ordinary software and reserve model calls for work that needs them.
What exists
Retrieval, routing, scheduling, validation, and hash-linked event-ledger modules. This public overview intentionally omits internal account and customer information.
Evidence & boundary
Source exists for these mechanisms. No current autonomous operation, revenue result, benchmark calibration, or independent security certification is claimed.
Developer experienceInternal tool
A mobile window into project work
Keep project context within reach.
A small browser interface for project selection, persistent conversation history, and streamed command output on a mobile device.
Read the product decision
The problem
Checking development work from a phone often means losing project context or navigating a desktop-shaped interface.
The decision
Keep the interface focused on choosing a project, reading its history, and following output as it arrives.
What exists
A Node.js and WebSocket application with SQLite history and a mobile-oriented interface. It integrates with a coding tool; it is not a new foundation model.
Evidence & boundary
Source supports the described internal-tool scope. Current remote accessibility, adoption, and security assurance are not claimed.