— PROOF OF CONCEPT · SURVEILLANCE INTELLIGENCE

Vigilant.ai

See every frame, tag everything.
AT A GLANCE
  • CATEGORYSurveillance intelligence
  • MODULESDetect · OCR · Semantic
  • INPUTDrop an MP4
  • STATUSProof of concept

Drop an MP4 and turn raw CCTV into a searchable, structured intelligence archive. Vigilant.ai runs object detection, licence-plate OCR, and semantic captioning automatically — then lets you find any event in seconds by natural language, plate, or object class. Built by IndiaNIC.

80
01 / OBJECT CLASSES DETECTED
3
02 / NEURAL MODULES
1 FPS
03 / FRAME SAMPLING
Seconds
04 / TO FIND ANY EVENT
WHY IT'S DIFFERENT

Stop scrubbing footage frame by frame.

Finding one vehicle or event in hours of CCTV is manual and unscalable — and there’s no way to query video by meaning. Vigilant.ai fixes both, showcasing our computer-vision and AI/ML engineering.

Multi-modal in one pipeline

Object detection, plate OCR, and semantic captioning together — not three disconnected tools.

Search footage by meaning

“Red car near the gate” surfaces the right frames; you don’t need to know the plate or the timestamp.

Frame-accurate forensics

Every result deep-links to the exact moment, with live bounding-box overlays you can toggle.

Toggleable & non-destructive

Run only the modules you need, then reprocess older footage to add capabilities without re-uploading.

FRAME-ACCURATE FORENSICS

Ask in plain language, land on the frame.

Search by meaning, plate, or object class. Results deep-link to the exact moment with live bounding-box overlays, and a timeline shows you where the activity actually is.

red car near the gateSemantic · 2 hits
CAM 04 · 23:14:07
Person #3 · 94%
Car · 91%MH01 AB 1234
Intelligence timelineactivity density
Illustrative explorer — natural-language search, frame-accurate bounding boxes, and a clickable activity timeline.
THREE NEURAL MODULES

Detection, plates, meaning — toggle per video.

Three modules turn footage into intelligence; switch each on or off per upload to balance compute against depth.

01 · MODULE

Object detection & tagging

Detects and tracks objects across 80 classes — each tagged with class, confidence, bounding box, and a persistent track ID.

02 · MODULE

Licence-plate recognition

Vehicle detections trigger a two-pass OCR pipeline that locates the plate, sharpens it, and reads the text — validated against regional formats.

03 · MODULE

Semantic AI search

Frames are embedded and captioned, stored in a vector database, so operators search footage in plain natural language.

HOW IT WORKS

Drop it, tag it, investigate.

The whole loop — from a raw MP4 to a searchable archive — runs as a background pipeline with live progress.

01

Drop an MP4

Upload raw footage and choose which pipelines to run — tagging, OCR, semantic.

02

AI auto-tags it

The background pipeline samples frames, detects and tracks objects, reads plates, and embeds scenes — with live progress.

03

Search & investigate

Query by language, plate, or class; jump to the exact frame with overlays, then export the evidence.

THE PIPELINE

Five steps, run in the background.

Every upload follows the same deterministic pipeline as an asynchronous task, so long videos process without blocking the operator.

01
Upload & store

Footage streams to cloud storage in chunks, generating a CDN URL.

02
Frame sampling

Frames are sampled at 1 FPS and capped in width to optimise compute.

03
Object detection

Frames pass the chosen model; stable track IDs are assigned and written in batches.

04
Plate OCR

Vehicle crops trigger plate localisation and dual-pass character extraction.

05
Semantic embedding

Frames become vectors plus human-readable captions for natural-language search.

SEARCH & INVESTIGATE

A command center for your footage.

Once footage is tagged, operators query, seek, and export from a single forensic workspace.

Omni-search bar

One query matches exact plate strings and semantic scenes at once, with a sensitivity slider and a global / per-video scope lock.

Interactive detail explorer

A deep-linked player with toggleable segmentation masks, colour-coded by class, and confidence-labelled track IDs.

Intelligence timeline

A 72-bucket activity-density waveform — taller bars mean more happening; click any bar to seek straight to it.

Insights & export

Objects, Plates, and Semantic tabs with filtering, low-confidence “manual check” flags, and a one-click CSV evidence export.

BUILT RIGHT

A real build, not a mockup.

Vigilant.ai is a working proof of concept from our team. Want it pointed at your camera network — or a different vision problem? Start with a rapid POC or browse the rest of the portfolio.

YOLOv8 / RT-DETRByteTrackEasyOCRSigLIP + BLIPPostgreSQL + pgvectorFastAPINext.js
— PROVE IT ON YOUR FOOTAGE

Turn footage into findable intelligence.

Vigilant.ai started as a proof of concept. Tell us what you need to find in your video, and we'll stand up a working build on your footage — fast.

— WHEREVER YOU ARE
hello@indianic.comWhatsApp Chat
RESPONSE TIME
< 4 hours
NDA
On request
FREE POC
3 – 5 days