AVANA TECHNOLOGY GROUP / SOFTWARE RESEARCH
OverTime.
The infrastructure for on-demand expertise.
Technical white paper · Draft 0.1 · September 8, 2026
OverTime Technology Group LLC · An Avana Technology Group project · Made with Sesame
Proposed architecture and product direction. This document describes what we intend to build, not a claim that these integrations, performance targets, regulatory processes, or payment capabilities are operational.
1. Human expertise as an available service
I’m building OverTime around a simple idea: valuable experience should have a direct path to someone who needs it. Professionals, retired specialists, and people with qualified, practical experience hold knowledge that can be useful long before a traditional appointment becomes available.
The ambition is a market for immediately available expertise—a cognitive liquidity market. Here, liquidity means making professional and experienced specialist time easier to discover, connect with, and compensate. It does not mean creating a financial security or tokenizing people. Corporate consolidation and automation motivate the project; the product hypothesis is that shorter, timely conversations can create another route to paid work.
OverTime is planned as a web application that would supply the connecting infrastructure: dispatch, acceptance, video connection, session records, and settlement. A consumer could request help directly. A business application or authorized AI agent could bring a specialist into an existing workflow through the same underlying platform.
2. A shared engine for direct and B2B access
- Request expertise
- Route to eligible experts
- Expert accepts
- Launch video
- Meter and settle
| Layer | Proposed responsibility |
|---|---|
| Client surfaces | A responsive web application for desktop and mobile browsers, business-app integrations, and AI-agent tools submit scoped consultation requests. |
| Gateway | Authenticate callers, enforce tenant boundaries, validate scopes and budgets, apply rate limits, and deduplicate requests. |
| Dispatch | Match expertise, credential eligibility, jurisdiction, language, current availability, and request context; notify a bounded set of experts. |
| Session orchestration | Resolve acceptance races, confirm readiness, issue short-lived room credentials, coordinate connection state, and publish lifecycle events. |
| Media | WebRTC signaling and connectivity services support the video consultation, with relay infrastructure when direct connectivity is unavailable. |
| Ledger and settlement | Reconcile session events, calculate charges, record platform fees, and instruct supported payment providers. |
Interactive video remains a client experience; the integration layer coordinates it. A third-party app can launch an authorized session surface or hand the user into OverTime rather than attempting to transport video through an AI tool call.
3. Third-party applications, AI, and MCP
The proposed API and Model Context Protocol (MCP) adapter expose one controlled service layer. For example, a business workflow could encounter an accounting question, ask permission to connect an expert, and request a consultation with a budget and a concise, approved context summary.
| Illustrative operation | Purpose |
|---|---|
expertise.request | Submit specialization, context reference, tenant, consent record, maximum rate, and spending limit with an idempotency key. |
consultation.status | Return request state and the authorized next action; never expose another tenant’s session. |
consultation.join | Return a short-lived, participant-bound launch URL after eligibility and authorization checks. |
consultation.cancel | Cancel an eligible request with a durable cancellation event. |
consultation.receipt | Retrieve an authorized statement of session time, charges, and settlement status. |
These are proposed contracts, not published endpoints. Partner webhooks would carry signed lifecycle events with event IDs, replay protection, retry handling, and versioned schemas. Consumers must handle duplicate and out-of-order events.
An AI agent should not have unrestricted authority to spend, share confidential files, or establish a professional engagement. Tenant policy and explicit user authorization determine allowable actions. Context sharing should be minimal, reviewable, and revocable. Retrieved documents and agent-generated content remain untrusted inputs.
4. From acceptance to live video in seconds
The core experience targets an in-platform video session launching within seconds of an expert accepting. This is a design objective to measure, not a demonstrated service-level guarantee. It depends on participant readiness, permissions, connectivity, and required pre-session checks.
Preflight should happen before dispatch wherever possible: confirm camera and microphone permissions, display the rate and spending cap, establish scope and consent, and verify expert eligibility. On acceptance, a transactional state change reserves one expert and releases competing offers. The orchestrator then creates the room, issues participant-bound credentials, and starts signaling immediately.
The proposed state sequence is requested → offered → accepted → connecting → live → ended → reconciled, with explicit timeout, cancellation, rejection, and failure paths. Acceptance alone must not count as a live session. Media-ready events from both participants establish the connection state, while server-side timestamps support reconciliation.
Measure acceptance-to-room-ready, acceptance-to-first-remote-frame, connection success, reconnect duration, and failed-session rates at median and tail percentiles. A disconnected user needs a clear retry or cancellation path; a failed connection must not silently become a completed consultation.
5. Metering, fees, and financial rails
The current product proposal uses a centralized fee on completed sessions. Rates, platform fees, billing increments, caps, and interruption treatment must be disclosed before connection. A double-entry ledger would separate customer charges, provider receivables, platform revenue, refunds, and payment-provider costs.
Session events require unique identifiers and idempotent processing. A monotonic duration source and reconciled media-state intervals should drive billing rather than trusting a browser timer. Sub-second measurement is distinct from the billing increment presented to customers.
Fiat and USDC are proposed settlement options, subject to payment-provider support, geography, onboarding requirements, and reviewed operating rules. USDC settlement introduces network selection, address validation, finality, and reversal limitations. Recording a payout instruction is not the same as confirming funds received. The choice of custody and funds flow remains an implementation decision requiring review.
6. Trust, eligibility, and data boundaries
Eligibility should reflect the help offered: relevant experience and demonstrated qualifications, with licensing checked where required. Credential checks should capture source, verification time, expiration, jurisdiction, and any review exceptions. An eligible credential does not establish suitability for every request. Scope, professional-client relationship requirements, and local restrictions must be defined for each supported field before launch.
The proposed guardrail layer coordinates reviewed notices, consent, confidentiality terms, and engagement requirements. Automatically generated disclaimers are not a substitute for professional or legal review.
- Use least-privilege tenant and participant authorization on every request.
- Keep media credentials short lived and bound to the intended session and participant.
- Minimize sensitive context and define retention, deletion, and access-audit policies.
- Treat recording as a separate consented capability; do not assume sessions are recorded.
- Protect webhook delivery, payment events, and credential updates against replay and duplicate processing.
- Maintain clear escalation paths for complaints, disputed charges, impersonation, and inappropriate conduct.
7. What must be proven
The first technical milestone is one complete request-to-video journey with real readiness checks, acceptance-race handling, connection failures, session reconciliation, and a sandbox payment flow. The first market milestone is demonstrating repeat demand for timely expertise and sustainable availability from qualified specialists.
- Measure connection speed and reliability on mobile networks and typical business devices.
- Test tenant isolation and scoped agent permissions, including adversarial inputs.
- Validate demand, specialist willingness, matching quality, and cancellation behavior.
- Evaluate pricing transparency, fee economics, dispute handling, and provider onboarding.
- Confirm field-specific requirements before offering regulated professional services.
B2B integrations extend the same engine into the software businesses already use. The long-term ambition is to provide the pipes for a market where human expertise is reachable at the moment it becomes valuable.