Veraxion extracts check-worthy claims from live speech in seconds, verifies them against
official
statistics and global fact-checking databases, and gives your newsroom human-approved
context
overlays for live broadcasts.
Real-time claim extraction, normalization, verification, and automated lower-third cueing.
Verifies numeric claims directly against embedded official Indian government snapshots (MoSPI, AISHE, MNRE).
100% Human-in-the-Loop newsroom safety. Nothing airs without explicit producer approval.
Never guesses or hallucinates. Explicitly outputs "Unverified in time" when evidence is inconclusive.
Instant broadcast kill switch clears lower-third overlays in under 500ms with a hotkey.
Why modern newsrooms and debate feeds require sub-second deterministic verification.
Cascading verification pipeline optimized for speed, compute cost, and strict journalistic accuracy.
Instant semantic matching against pre-verified claim vectors for sub-second lower-third generation.
Queries Google Fact Check Claim Search and recognized global fact-check organization repositories.
Executes deterministic SQL queries against embedded SQLite government databases (MoSPI PLFS, CPI, AISHE, MNRE).
Local FAISS vector search combined with Tavily web retrieval and MiniCheck natural language inference scoring.
Automatically flags unresolved claims as "Unverified in time" and routes to human editor triage.
Dedicated interface views tailored for each member of the broadcast production team.
| User Role | Primary Screen | Workflow Action | Value Delivered |
|---|---|---|---|
| Broadcast Operator | /console |
Monitors wire items, inspects evidence, approves overlays | Cuts fact-checking triage from 20 mins to 5s per claim |
| Debate Moderator | /moderator |
Receives short, data-backed follow-up questions on tablet | Enables anchors to challenge vague stats live |
| Broadcast Audience | /overlay |
Views neutral lower-third overlay graphics on delayed feed | Builds deep trust without interrupting show flow |
| Station Executive | /results |
Audits past sessions and SHA-256 evidence logs | 100% editorial compliance and auditability |
Why leading networks choose Veraxion over manual checking or generic LLM tools.
| Feature | Manual Checking | Generic LLMs | Veraxion Platform |
|---|---|---|---|
| Verification Speed | 15–60 minutes | 5–15 seconds (unreliable) | 1–3s (Cached/Official), Max 30s |
| Live Delay Sync | Impossible | High hallucination risk | OBS Overlay with 30s Delay Sync |
| Official Data Engine | Manual lookup | Guesses figures | Deterministic SQLite Computation |
| Air Safety Controls | Shouting / Phone | Uncontrolled API risk | Human Operator Desk + 500ms Kill Switch |
| Neutrality Standard | Subjective tone | Unpredictable prompt | Neutral Evidence Cards (No True/False Labels) |
High-performance zero-build stack powered by Python FastAPI, WebSockets, Deepgram Nova-3, and GLiNER.
A: All primary buttons redirect directly to the official Veraxion GitHub repository:
https://github.com/VishnupiriyanV/Veraxion.
A: Veraxion enforces a strict Human-in-the-Loop design. No graphic appears on screen automatically. An operator must explicitly review the evidence card and click "Approve for Air."
A: The system routes the claim through Tier 2 vector search and allowlisted web retrieval. If evidence remains inconclusive before the 30s deadline, it outputs "Unverified in time."
A: Veraxion processes undelayed live audio and cues approved lower-third graphics to display on the delayed feed at the exact second the speaker's claim reaches home viewers.