AI Fact-Checker Overlay · Android · iOS coming

VeRa

One swipe. One truth.

The AI fact-checker that lives over any mobile app. No copy-paste. No app switching.

VeRa (from Latin veritas — truth) is a personal AI fact-checker that lives as an overlay on top of any app on your phone. Spot a suspicious story on Facebook, X, Instagram, or any news site — pull the lens over it, and within seconds get a verdict with citations to real sources.

For investors
ProductAI Fact-Checker Overlay
PlatformAndroid live · iOS in progress
StatusWorking APK · Demo live
↓ scroll

01 — The Problem

We live in an information storm.

Misinformation spreads faster than truth because verification requires effort — and no one makes that effort mid-scroll.

Every day, billions of posts, headlines, and videos compete for attention — and the noise is winning. AI-generated fakes are indistinguishable from real reporting. Manipulated content spreads faster than corrections. Outrage travels at the speed of a share button, while the truth is still loading.

The result: most people can no longer tell what's real. Not because they're naive — but because the volume of noise has made reality structurally hard to see.

38%

Americans regularly get news on Facebook

35%

get news via YouTube video

57%

of X users get news there

20%

get news on TikTok (up from 1% in 2020)

For the first time in 2026, social media surpassed TV as the leading news source globally — Reuters Digital News Report 2026.

The core problem: verifying a claim means leaving the app, Googling, cross-referencing sources — a process almost no one completes mid-scroll. Misinformation gets shared before anyone checks it.


02 — Why Now

Trust is at a low. Tools haven't caught up.

Social media surpassed TV

For the first time in 2026, social media became the leading news source globally (Reuters 2026). Verification needs to happen where people read — not after.

AI fakes are now indistinguishable

Anyone can generate convincing fake articles, images, and videos in seconds. The volume of unverifiable content is growing faster than anyone can check it.

Trust in media at historic lows

Reuters Institute and Edelman Trust Barometer (2025) record audience trust at its lowest in two decades. People are looking for an alternative — but don't know what to trust.

EU AI Act tightens the rules

Regulators are demanding transparency from AI products. For VeRa, this is already architecture — not compliance work to do later.


03 — Solution

Verification in one gesture.

VeRa is a system overlay. Pull the lens over any content on screen, and get a structured verdict with cited sources — without leaving the app you're reading in.

01

Swipe the handle

A small handle sits at the edge of your screen. Swipe it open whenever something on screen looks off — in Facebook, X, Instagram, WhatsApp, a news app, anything.

02

Frame the claim

Drag the lens over the content you want to verify. VeRa captures the frame on-device, extracts the text via on-device OCR, and never sends the image anywhere.

03

Read the verdict

Seconds later, a card appears with one of four verdicts — APPROVED, FAKE, MANIPULATION, or UNVERIFIED — plus the key analysis and links to the sources VeRa actually used.

No copy-paste. No app switching. The screenshot never leaves your device.


04 — Four Verdicts

One system. Four outcomes.

Every verdict comes with a key analysis, cited sources, and a confidence value — never an opinion, never a fabricated answer.

Why orange for FAKE and not red? Red carries strong political and partisan associations in many markets. VeRa is neutral by design — orange signals alert without taking a side.

✓ Approved VeRa

The Stockholm3 blood test is a validated diagnostic tool confirmed by NICE-referenced studies.

Key Analysis

External evidence confirms Stockholm3 combines genetic markers, proteins, and clinical data.

Sources & Evidence

⛓ Stockholm3 | The Doctors Laboratory

⛓ Stockholm3 — EIT Health

Confidence 95%
⊗ Fake VeRa

The claim that extraterrestrials are unveiling technology is a fabrication with zero factual basis.

Key Analysis

No evidence supports the existence of alien entities or their involvement in terrestrial AI development.

Sources & Evidence

⛓ No independent sources found

Confidence 100%
⚠ Manipulation VeRa

The article uses a sensationalised headline to misrepresent policy disagreements as the root cause of criminal acts.

Key Analysis

The framing creates a false causal link between policy and violence not supported by evidence.

Sources & Evidence

⛓ Man Charged After Five People Injured In Suspected Hate Crime

⛓ 5 men stabbed in alleged anti-Muslim attack — UPI.com

Confidence 95%
? Unverified VeRa

The claim cannot be confirmed — insufficient independent sources were found.

Key Analysis

The input provides only a generic headline without specific facts to verify.

Sources & Evidence

⛓ No independent sources found

Confidence 30%

These are real outputs from the VeRa engine, presented as illustrative examples. They are not a measured accuracy claim.


05 — In Context

Lives over any app.

The overlay floats above whatever the user is reading — Facebook, X, Instagram, WhatsApp, any news site. One swipe. Verdict in seconds. No app switching. No copy-paste.

✓ Approved ⊗ Fake ⚠ Manipulation ? Unverified
Try the live demo →

06 — Differentiation

Why VeRa is structurally different.

Five properties that browser extensions, fact-check websites, and general-purpose chatbots cannot replicate inside a mobile app.

01

In-context overlay UX

Works over any mobile application without leaving context. Browser extensions don't function inside native apps. Websites require a context switch. No equivalent exists in the mobile space.

02

Privacy by design

Capture and OCR happen on-device. Only text leaves the phone — never the image. A few hundred bytes per check, no biometric data, nothing personal.

03

Verdict integrity by architecture

No sources → UNVERIFIED. Never a fabricated answer. RAG with citations eliminates the kind of hallucinations that plague general-purpose chatbots.

04

Source-transparent

Every verdict is built on public sources via RAG. Each citation is open — anyone can verify it independently. No need to take our word for it.

05

Language-mirroring

The engine responds in the language of the content automatically. No setting to toggle, no UX friction.


07 — Architecture

Privacy is not a policy. It is the architecture.

VeRa is built on a strict separation between what stays on your device and what reaches our servers. The screenshot never leaves your phone. Text — and only text — is transmitted.

On the device

  • Flutter overlay (system layer)
  • Android MediaProjection (frozen frame)
  • On-device OCR via Google ML Kit
  • Image processed and discarded locally

In transit

  • Only OCR text is sent
  • Typical request size: ~300 bytes
  • Authenticated, rate-limited, TLS
  • No image, no biometrics, no personal IDs

On the server

  • Real-time RAG via Tavily web search
  • LLM verdict via Ollama + Qwen3.6
  • Live SSE stream: extraction → search → verdict
  • Verdict cached locally; no training on user input

The system is architecturally incapable of "confidently lying" — this is a technical constraint, not a behavioural policy. Where regulators ask for AI transparency (EU AI Act and others), we already meet the bar by design.


08 — Where we are

Working product, not a deck.

LIVE

Signed Android APK · end-to-end flow confirmed

117 / 0

Backend tests passing · health 10/10

~$10

Annual infrastructure cost (domain only — Ollama on existing hardware, Cloudflare Tunnel free tier)

OPEN

Architecture audit: OWASP / STRIDE — no leaks or injections

iOS

Planned. Multi-language OCR on roadmap.

Demo is live at redmindsys.uk — fully working end-to-end pipeline on a single PC. The next step is cloud inference, iOS, and broader OCR coverage.


09 — AI Ethics

Honest by construction.

The entire data pipeline is built on a small set of non-negotiables.

Capture

Screenshot never leaves the device. Only text is transmitted.

Search

RAG retrieves real, public sources. Nothing is fabricated.

Verdict

No sources = UNVERIFIED. Sources are always shown to the user, never hidden.

Data

We do not train on user queries and do not scrape third-party content. Pretrained models + RAG only.

"Assistive, not authoritative." VeRa helps you reach the right conclusion — it does not declare it for you.


10 — For Investors

We are raising — and we are open to multiple paths.

VeRa is at the rare stage where the product works, the cost base is near zero, and the next step is scale: cloud inference, iOS, multi-language OCR, and the first marketing campaign.

Pre-seed angel / VC

For partners who want to take a position in an early, technically clean AI infrastructure play.

Non-dilutive: grants & accelerators

EIC, Techstars, MassChallenge, Startup Wise Guys, Google for Startups Ukraine — programmes we're evaluating in priority order.

Crowdfunding

Kickstarter or Indiegogo — both technically compatible with our AI disclosure stance (Consent & Credit 2024+).

Contact

If you cover AI, mobile, infrastructure, or trust-and-safety — we'd be glad to talk.


11 — FAQ

The questions we hear most.

Why not just use ChatGPT? +

ChatGPT requires copy-paste, context switching, and doesn't cite real sources for verdicts. VeRa is purpose-built: one gesture, on-device OCR, RAG with real sources, structured verdict.

How do you handle legal risk? +

Verdicts are based on public sources via RAG, never generated. When evidence is insufficient, the verdict is UNVERIFIED — never a fabricated answer. We use "assistive, not authoritative" framing. The primary protection is architectural, not contractual.

What's the AI disclosure story? +

Pretrained models + RAG. No training on scraped data. Sources are always cited. The system is architecturally incapable of fabricating a verdict.

What's the moat? +

On-device OCR (privacy) + in-context overlay (UX no competitor has on mobile) + RAG with evidence-based verdicts (trust) + language-mirroring.

Where does the product run today? +

Signed Android APK, live at redmindsys.uk. End-to-end flow confirmed. 117 backend tests passing, health 10/10.

What is the constraint today? +

Inference runs on a single local PC (Ollama + Qwen3.6, 24GB). Demo mode, not production scale. Cloud inference and iOS are the next steps.


12 — Get in touch

Tell us who you are.

Investors, accelerators, journalists, partners — drop a line. We read every message.

We don't share your details. No newsletter, no tracking pixels. Just an actual human reading your message.