VeRa appears as an overlay on any application. One gesture → text extracted → sources verified → verdict in seconds. No context switching. No copy-paste.
An AI overlay for real-time content verification. Works on top of any application on Android without switching between windows.
Disinformation, manipulation, and fabricated narratives are a systemic feature of the information environment — not an exception. Social media, news, business communications, legal evidence: no channel has a built-in verification mechanism.
The pain materialises at two distinct moments:
You do not want to be deceived. Consuming false information means making decisions — financial, reputational, legal — on false premises. Without a tool, a person cannot reliably distinguish fact from manipulation.
You do not want to deceive others. Forwarding a fabricated story destroys trust with clients, audiences, and colleagues. The reputational risk is realised at sharing — but its root cause is the moment of consumption.
Existing verification tools require manual app-switching. That is why most people do not use them — the friction is too high. VeRa removes this barrier entirely.
Sources: Pew Research Center, Social Media and News Fact Sheet (Sept 2025); Ofcom Online Nation 2025.
Two forces converged in 2025–26. The cost of producing fabricated content collapsed: a convincing fake article now takes seconds to generate with commercially available AI. At the same time, the cost of verification fell equally fast: LLM inference costs decline roughly 50% per year, putting real-time claim-checking within reach of a mobile application.
The window opened now: disinformation at industrial scale + verification at marginal cost → for the first time, the economics of truth-checking work on consumer hardware. A 2013 case study puts the stakes in context: a single fabricated tweet from the AP's compromised account erased approximately $136 billion in S&P 500 market capitalisation in minutes. That was with a reach of thousands of followers. Today the same fabrication reaches millions in seconds, via platforms where VeRa operates natively.
Social media surpassed television as the leading news source in the US for the first time in 2025 (Reuters Institute DNR 2025). Verification needs to happen where people read — inside the app, in one gesture.
Anyone for whom truth matters: traders, journalists, content creators, lawyers, active media consumers. The defining characteristic is not sharing behaviour — it is the need for verification: establish what is true first, then decide what to do.
Priority launch segment: retail traders. A single viral fabrication converts directly into measurable financial loss. The motivation to verify is immediate and quantifiable — no other segment has this property at the same intensity.
One gesture initiates the full pipeline: extraction → search → analysis → verdict.
The verdict card appears as a floating overlay over any open application. Colour-coded status, confidence percentage, key finding, and cited sources — all within the reading context, no app-switching required.
Colour palette: APPROVED green · MANIPULATION amber · FAKE orange (orange is deliberate — red carries partisan connotations; orange signals alert without political association).
Engine: rin-v5.3 — includes temporal grounding, news-existence calibration, and language-mirroring (verdict in the language of the content; EN/RU/DE tested).
⬇ Download Android APK · redmindsys.uk| Layer | Technology | Function |
|---|---|---|
| Client | Flutter + Kotlin | Android overlay (MediaProjection), cross-platform UI |
| OCR | Google ML Kit | On-device — screenshots never leave the phone |
| Source retrieval | Tavily | Real-time search, verified sources, RAG-compatible |
| AI model | Qwen3 + Ollama | Claim analysis, verdict synthesis (rin-v5.3) |
| Backend | Node 24 | Streaming (SSE), verdict cache, 122 tests |
| Tunnel | Cloudflare Tunnel | Live HTTPS deploy, tester distribution |
A single data structure for any content type — the same interface will serve text, voice, and video, enabling B2B integrations to be built once.
| Field | Content |
|---|---|
| status | APPROVED / FAKE / MANIPULATION / UNVERIFIED |
| confidence | 0–100% — model confidence, shown to user |
| verdict_line | Single-line summary |
| key_fact | Primary fact on which the verdict rests |
| analysis | Detailed claim analysis |
| trigger_points | Specific elements that fired as signals |
| evidence[] | Sources: URL, headline, quote |
Reliability: zero sources from RAG → verdict automatically downgrades to UNVERIFIED; high confidence without evidence is architecturally impossible. Internal eval harness with regression cases tracks verdict quality across every engine change. A labelled-corpus benchmark is the first data milestone post-launch.
Active now: Android for traders. Funded by this raise: iOS and Accessibility. Voice, video, and spatial computing are the vision for the next product cycle, not current commitments.
Timeline milestones are counted from round close.
Android overlay on any application. Swipe → on-device OCR → Tavily source retrieval + Qwen3 → verdict in seconds. 122 backend tests, 0 regressions. Signed APK, live tester distribution at api.redmindsys.uk/download.
The same mechanic (swipe → verdict) extends to new content types. One engine, one data contract — new surfaces. This is a direction, not a dated commitment.
VeRa = veritas = truth. One name, one narrative — each audience reads it through its own context.
Competing solutions require manual switching between applications. VeRa is embedded in any context via a system overlay. The barrier to verification is one gesture — and a year of native edge-cases is not trivially reproduced.
Screenshots stay on-device; only extracted text is processed downstream. This minimises the legal surface regardless of jurisdiction and is an architectural decision, not a compliance label.
Internal eval harness with regression cases across every engine change. Zero sources from RAG → UNVERIFIED, never a high-confidence fabrication. This is a process that distinguishes VeRa from a raw API wrapper.
Each opt-in verdict enriches the training dataset → fine-tuning → improved accuracy → broader audience → more data. The competitive advantage compounds with scale. Currently empty; activates with users.
| Who | What they do | VeRa's distinction |
|---|---|---|
| NewsGuard | Trust ratings for news sites; browser extension / enterprise licences | Rates the source, not the specific claim; desktop-first; not in the mobile reading flow |
| Ground News | Coverage comparison and bias analysis across outlets | Perspective tool, not true/false verdict; requires opening a separate app |
| Platform fact-checks (Gemini, Copilot) | Claim checking inside their own chat interface | Does not operate over third-party apps (Telegram, TikTok, X); the platform is an interested party |
| Fact-check sites (Snopes, AFP Fact Check...) | Manual, editorial verification published after the fact | Hours to days; requires leaving context; cannot verify novel claims in real time |
| VeRa | Per-claim verdict over any app, one gesture, cited sources, seconds | In-context · per-claim · platform-neutral · mobile-native · privacy by design |
| Model | Segment | Mechanic |
|---|---|---|
| Freemium | B2C | 5–10 daily checks + last 5 verdicts free · unlimited checks & full history ~$6–8/mo |
| API / SDK | B2B — journalists, traders, lawyers | Programmatic interface embedded into enterprise workflows; ~$0.05/check |
| Infrastructure layer | Platforms, media groups | Anti-disinformation as a managed service · long-term vision |
VeRa is the first product of Redmindsys, an AI studio building intelligent verification infrastructure.
One founder, AI agents as engineering leverage, a growth advisor. What previously required five engineers and six months was shipped in weeks — nights and weekends.
No hires, no cloud spend, no external contractors:
"Isn't a solo founder a risk?" — Look at the output, not the org chart. The product in this deck was shipped by one person orchestrating AI agents — no hires, no cloud spend. The model scales: funding converts the founder to full-time on day one and adds two targeted hires (backend/AI engineer, B2B sales). Clean cap table: founder-held, advisor allocation reserved.
| Role | Priority | Why now |
|---|---|---|
| Backend / AI engineer | First hire | Cloud infrastructure migration, inference pipeline, eval corpus build |
| B2B sales / partnerships | First hire | Prop firm outreach, first enterprise contracts |
Verified prototype. Defined business model. This raise funds the transition from demo to production infrastructure and the capture of the first paying segment.
These are modelled figures, not billed actuals — no paying user base yet. Search and inference provider prices verified July 2026 (Serper $0.0003–0.001/query · Tavily $0.008/search · Gemini Flash-class and Groq Llama-class inference well under $0.002 per verdict at 3–4K tokens); re-check before any investor meeting — prices move every 6 months, typically downward. B2B API modelled at $0.05/check → 80–90% gross margin.
Economics improve with scale: cache hit rate grows with audience; model inference costs fall ~50%/year industrywide — no product change required.
| Allocation | Objective |
|---|---|
| Production infrastructure | Cloud inference migration (Groq/Flash), backend scaling to production load |
| iOS release | Screen Recording Extension port, App Store — same engine, adapted gesture |
| Team | First backend/AI engineer + first B2B sales manager |
| Research | Feasibility assessment for Voice and Video detection — foundation for next product cycle |
| Go-to-market | Trader community acquisition, first B2B pilot conversations, CAC measurement |
Prototype: working Android APK, 122 tests, signed build, live tester distribution.
Business model: defined. Freemium + B2B API with a documented path to revenue.
Legal entity: incorporation in progress (England & Wales, filing 6 July).
Accelerators: Applying — Y Combinator (Fall 2026) · Hub71 Abu Dhabi (Cohort 20) · Antler UK (London residency).