PublicVoiceShieldComment intelligenceAbout & Contact

Development history

PublicVoiceShield evolved through product research, problem definition, rapid implementation and iterative refinement. The milestones below are grouped by month rather than presented as a commit-by-commit timeline.

Pre-development concept phase

  • Identified the impact of hostility and trolling on public participation
  • Defined the need to distinguish criticism, disagreement, abuse and genuine questions
  • Established the principle: “AI advises. Humans decide.”
  • Developed the idea of serving public figures, organisations, supporters and members of the public
  • Outlined the responsibilities of handling public commentary safely and lawfully
  • Set out the expectation that every stored record would be tenant-isolated from the outset

June 2026

  • Early product research and decision-framework development
  • Exploration of public-figure communication, audience risk and responsible reply support
  • Definition of human oversight, escalation and audit requirements
  • Data modelling work: deciding what needed to be stored, for how long, and who should be able to see it
  • Design of the multi-tenant database structure around organisations and their members
  • Planning of the append-only audit approach so decisions could be evidenced rather than overwritten

July 2026

  • PublicVoiceShield MVP implementation
  • Database installation and configuration on managed PostgreSQL
  • Creation of the core schema: profiles, organisations, organisation members and invitations
  • Creation of the analysis schema: social posts, thread analyses, analysed comments and comment decisions
  • Creation of the research and evidence schema: research sessions, queries, sources, cache and evidence approvals
  • Creation of the governance schema: privacy requests, policy versions, policy acceptances, cookie preferences and retention settings
  • Row-level security policies, explicit table grants and security-definer membership checks
  • Database indexing and performance review for higher comment and thread volumes
  • Authentication and organisation accounts
  • Analyse Thread workflow
  • Secure server-side AI analysis
  • Reply identity modes
  • Human approval and audit trail
  • Comment Review workspace
  • Optional public-source research
  • UK-law-assisted replies
  • Privacy, legal and compliance pages
  • Retention wording corrections
  • Security and scalability review
  • Tenant-isolation verification across organisations
  • Tutorial video
  • Private source-control integration
  • Published controlled MVP

July 2026 — Professional Email, Support and Privacy Communications

  • Introduced a branded email and case-communication system across authentication, support, privacy and security workflows
  • Added fixed server-side sender identities for automated account emails, customer support, privacy rights requests and security correspondence
  • Sender selection controlled by the server, so browsers and users cannot supply arbitrary sender addresses, reply-to addresses or recipients
  • Branded acknowledgements for support cases, with customer-visible replies and closure messages remaining subject to authorised human action
  • Internal investigation notes kept separate and never included in customer emails
  • Structured PublicVoiceShield reference numbers for privacy rights requests
  • Controlled acknowledgement, progress-update and outcome workflows for privacy correspondence
  • Restricted delivery audit log recording sender identity, request reference, delivery status, approver and provider message information
  • Professional PublicVoiceShield mailboxes configured for support, privacy and security correspondence on the organisation's own domain

July 2026 — Platform Assurance and Workspace Refinements

  • Independent security review of database access controls, with least-privilege permissions applied to restricted administrative records
  • Review of third-party software advisories, with findings assessed, documented and recorded as part of ongoing maintenance
  • Further hardening of assistant and support records so access remains limited to the owning account or authorised platform staff
  • Cost and performance review of the AI processing pipeline, with usage recorded against each analysis for internal oversight
  • Added the ability to attach later comments to an existing analysed post, with a preview and confirmation step before anything is saved
  • Refinements to comment separation and delivery confirmation within the comment-addition workflow
  • Tidied the global footer and relocated contact details into the relevant support, privacy and security areas

July 2026 — Mobile Readiness and Brand Consistency

  • Readiness assessment for a mobile application build, covering layout, navigation and account journeys on smaller screens
  • Mobile presentation hardening, including safe-area handling and touch-friendly controls throughout the workspace
  • Added an authenticated account closure journey to meet mobile app store expectations
  • Prepared the supporting configuration and documentation required for a future store submission
  • Real-device testing on mobile browsers, covering sign-in, analysis, review and account journeys
  • Consistent PublicVoiceShield branding across browser tabs, home-screen icons and empty-state artwork
  • Upgraded the published plan so no third-party editing badge appears, giving users a fully PublicVoiceShield-branded experience

July 2026 — Controlled Beta Access and Cost Governance

  • Introduced approved-access controls so the service runs as a controlled beta rather than an open sign-up
  • Server-side usage allowances for AI-assisted work, applied per account and enforced before any request is processed
  • Usage recorded only against successful results, with allowances returned automatically when a request does not complete
  • A protected primary owner role, with account roles and status shown in an administration view
  • Administration dashboard for reviewing accounts, approving access and monitoring activity and cost
  • Configurable warning levels and an overall processing limit, with a global pause control for AI-assisted features
  • Owner notifications for new sign-ups and threshold events, using the branded email identities
  • Introduced an optional voluntary contribution towards development, presented in the Help Centre with no effect on access or features
  • Registered interest recorded securely and relayed to the support mailbox for a human response

August 2026 — Controlled Beta Test Planning

  • Prepared a structured controlled-beta test plan covering eight assessment areas
  • Account access: sign-up, verification, sign-in and password recovery journeys
  • Organisations: creation, onboarding, member invitation and member removal
  • Analysis quality: separating disagreement, criticism, abuse and genuine questions
  • Reply support: identity and tone handling across the available response options
  • Ongoing threads: adding later comments to an existing analysed post
  • Evidence features: optional public-source research and UK-law-assisted replies
  • Governance: usage allowances, cost controls and audit records
  • Communications: support, privacy and security email journeys
  • Mobile: repeat testing of the full workflow on phone-sized screens
  • Researched and compared options for participant recruitment, feedback capture and issue triage
  • Recorded the preferred options, assessed against cost, privacy exposure and administrative effort

August 2026 — Assisted Decision Quality and Cost Discipline

  • Structured review of how assisted work is routed, so routine high-volume steps and more sensitive judgement steps are handled proportionately
  • Anonymised side-by-side evaluation against existing test material, measured on accuracy, relevance, tone, caution and responsiveness
  • Cautious-by-default behaviour reinforced for sensitive material, with human referral preferred over an automated suggestion
  • Per-request cost and usage recorded for internal oversight, with warning levels and an overall processing limit retained
  • Owner-only view for reviewing routing decisions and evaluation results

August 2026 — Access Control Assurance

  • Independent review of account permissions across invitations, onboarding and profile updates
  • Server-side enforcement of role assignment, so no permission can be granted from the browser
  • Database-level guards added alongside application checks, giving two independent layers
  • Regression tests covering privilege escalation attempts, retained as part of the standard test suite
  • Repeat security scan completed after the changes

August 2026 — Relevance by Organisation Type

  • Introduced an organisation-type-aware catalogue so each account sees only the optional workspaces relevant to it
  • Types covered include elected representatives and public offices, councils and local authorities, housing associations, retail and customer-service organisations, charities, education, healthcare, public bodies and communications agencies
  • Unreleased areas are shown as planned rather than offered, and are never switchable by an account owner
  • Every eligibility rule re-applied on the server, with the browser view treated as presentation only
  • Automated tests covering every organisation type, to confirm the offered set changes with the selection

August 2026 — Informed Onboarding and Governance Controls

  • Removed any pre-selected organisation type so the choice is always deliberate
  • Each type presented as a described choice that states plainly what it changes and what it does not
  • Settings mirrors the onboarding wording, so the two never disagree
  • Organisations we have not yet classified are marked for review rather than granted anything automatically
  • Platform-owner governance controls for correcting an organisation type or recording a documented exception, with a reason required and an audit record kept

August 2026 — Working at Scale

  • Account administration reworked into a searchable, sortable and collapsible list with status filtering and organisation context
  • Direct links from an account to the relevant governance controls
  • Comment review toolbar for long discussions, with ordering by sequence, author, risk or undecided-first
  • Filtering by category and recommended action, with progressive loading for very long threads
  • Internal notification when a new support case is raised, so requests reach a person promptly

August 2026 — Audience Understanding Workspace

  • First optional workspace released, grouping already-analysed comments into recurring themes
  • Recurring unanswered questions surfaced, with the human decision on any clarification recorded
  • No new data collected: the workspace reuses material the organisation has already submitted
  • Wired into the controlled-beta controls: approved accounts only, usage allowances, cost limits and the global pause
  • Existing accounts brought onto the same allowance basis as new ones

August 2026 — Privacy by Design and Records of Processing

  • Diagnostic logging rewritten so personal details and access tokens are removed before anything is recorded
  • Account closure extended to clear the settings and allowance records that can be safely removed, with the limited items kept for audit stated openly
  • Settings and closure screens updated so people are told what is deleted and what is retained, and why
  • Supplier and processing records corrected to describe the actual route taken by assisted analysis
  • Formal documentation pack drafted for the operator's compliance file, with outstanding owner decisions listed rather than assumed

August 2026 — Public Site Quality and Discoverability

  • Independent read-only review of the live site covering links, response codes, page titles, descriptions and mobile usability
  • Confirmed findings corrected: canonical and social addresses now point to the live domain on every public page
  • Search-engine instructions and the site index aligned to the same domain, with sign-in and account screens excluded from indexing
  • Touch targets on shared page elements enlarged to meet accessibility guidance
  • No page wording, design, access rules or product behaviour altered by the corrections

August 2026 — Assurance of Controlled Access

  • Re-confirmed that every new sign-up is held for the operator's approval before any assisted feature can be used, enforced on the server rather than in the interface
  • Administrative registers restricted so sensitive governance references are visible only to the operator, with a reduced view for signed-in users
  • Governance trail confirmed as append-only and unreachable from the application
  • Independent security review re-run, findings addressed or recorded with reasons

August 2026 — Multi-Platform Link Recognition

  • Support for links from the major public discussion platforms, recognised by address only
  • No account connection, no automated collection and no posting, hiding or reporting anywhere: discussion text is always supplied by the person using the service
  • Analysis records reduced to structure and outcome only, with no capacity to hold submitted wording
  • A clear control to clear the working session, with plain wording about what that does and does not cover
  • Operator-only register recording the permitted status of each platform, with every change logged and a reason required

August 2026 — Processing Route Review

  • Reviewed how assisted analysis reaches its processing provider, and prepared a simpler, more directly accountable route
  • Owner-only test mode used first, so nothing changed for accounts while the option was assessed
  • Anonymised side-by-side comparison against existing test material before any recommendation was made

August 2026 — Supplier Assurance and Approved Wording

  • Supplier data-processing terms reviewed and formally recorded, including retention position and applicable safeguards
  • Every proposed public wording change presented side by side for owner approval before publication
  • Amendments applied exactly as approved, with unverified details marked as unconfirmed rather than assumed

August 2026 — Controlled Change and Verification

  • Agreed a strict sequencing rule so published wording never described a route that was not yet in use
  • Change carried out as a single controlled operation: readiness checks, publication, switch, verification and a prepared rollback
  • No substitution behaviour: a request that cannot use its intended route is refused and referred for owner review rather than quietly handled elsewhere
  • Every step recorded in an append-only governance trail
  • Read-only pre-publication and post-publication verification against the live site and live settings

August 2026 — Presentation Refinement

  • Design pass for a calmer, more authoritative feel, within the existing brand identity
  • Clearer hierarchy, steadier spacing, refined panels, tables and empty states, and a clearer distinction between ordinary and higher-risk actions
  • Improved readability on long governance pages and further mobile refinement
  • No change to wording, permissions, routing or product behaviour

Created and developed by Mohammad Asif. Built from sustained product thinking, rapid implementation and iterative testing.