Sovereign PortfolioPolice & Public SafetyAI Police Filing (NPA Project)
Winner: BHASHINI LEAP Hackathon (Law Enforcement PS2)SVP National Police Academy / NCRB Ecosystem

AI-Driven Voice-to-FIR & BNS 2023 Automation Platform

Transforming distressed citizen speech into structured, legally verified First Information Reports in under 90 seconds. Record → Verify → File.

“Selected as the Winner of Problem Statement 2 in the BHASHINI LEAP Hackathon. Freeing investigating officers from 4–6 hours of repetitive shift documentation while ensuring 100% legal precision under the newly enacted Bharatiya Nyaya Sanhita.”
< 90 Sec
FIR Draft Generation Time
Reduced from 45 minutes of manual slow keyboard typing by the IO.
17,000+
Target Police Stations (Thanas)
Nationwide deployment footprint across State & Union territory police forces.
4–6 Hours
Officer Shift Time Reclaimed
Hours spent daily typing FIRs and case diaries diverted back to field policing.
BNS 2023
Penal Code Legal RAG Alignment
Automated mapping from factual elements to Bharatiya Nyaya Sanhita clauses.
Operational Field Reality

17,000+ Thanas: The Documentation Bottleneck in Criminal Justice

4–6 Hours
Shift Typing Fatigue Per Officer
Authority: Bureau of Police Research & Development (BPR&D) Shift Workload Analysis

When a traumatized victim enters a police station, they narrate incidents in conversational regional language. An Investigating Officer (IO) must manually listen, take notes, convert colloquial slang into formal court terminology, and recall appropriate sections under the newly enacted Bharatiya Nyaya Sanhita (BNS 2023). This manual process takes 45 to 60 minutes per complaint, consuming up to half of an officer's shift in repetitive typing.

The BNS Transition Friction: Transitioning from 164 years of IPC sections to BNS 2023 creates recall confusion, incorrect section citations, and vulnerable charge-sheets during trials.
01

Trauma & Regional Dialect Slang

Citizens in distress narrate incidents in colloquial regional dialects and slang. Generic speech recognition systems fail completely on informal grammar, local weapon names, and rapid emotional speech.

02

Ambient Thana Acoustic Noise

Police stations are inherently noisy public environments with overlapping conversations, wireless radios, phone bells, and vehicular sirens. Off-the-shelf ASR engines produce fragmented, incoherent output.

03

IPC to BNS 2023 Penal Realignment

Officers trained over decades in IPC section numbers (e.g. 379, 392, 420) must now formulate FIRs under BNS clauses. Missing a critical statutory subsection can lead to bail loopholes in court.

Legal Realignment Matrix

IPC 1860 $\rightarrow$ Bharatiya Nyaya Sanhita (BNS 2023) Automated RAG Mapping

MultiLipi's Legal RAG engine automatically maps extracted incident facts to the exact corresponding BNS penal provisions:

Offence ClassificationOld IPC 1860 SectionBharatiya Nyaya Sanhita (BNS 2023)MultiLipi RAG Mapping Rule
Theft (चोरी)IPC Sec 379BNS Sec 303(2)Extracted missing moveable property without consent
Robbery with Hurt (लूट व चोट)IPC Sec 392 / 394BNS Sec 309(4)Physical force or weapon threat accompanying property seizure
Voluntarily Causing Hurt (मारपीट)IPC Sec 323BNS Sec 115(2)Bodily injury inflicted without deadly weapon
Criminal Breach of Trust (विश्वासघात)IPC Sec 406BNS Sec 316(2)Misappropriation of entrusted commercial/personal property
Cheating & Dishonesty (धोखाधड़ी)IPC Sec 420BNS Sec 318(4)Deceptive inducement causing financial or property delivery
End-to-End Operational Pipeline

The Record $\rightarrow$ Verify $\rightarrow$ File Law Enforcement Workflow

From noisy verbal narration to structured CCTNS export in under 90 seconds, preserving full Investigating Officer discretion at every juncture.

Workflow Stages (Click to Inspect):
Stage 01 of 07Acoustic Capture
CCTNS Native Pipeline

Complainant Spoken Intake

Dual-microphone directional intake isolates complainant's voice stream while suppressing background station clamor (sirens, phone rings, multi-party conversations).

Acoustic / Data Input:

Traumatized victim/witness speaking in conversational dialect (Hindi, Marwari, etc.) inside noisy thana

Structured Legal Output:

Calibrated 16kHz isolated primary speech audio stream with ambient noise suppression

Neural Model & System Specifications

Directional Beamforming • Silero Voice Activity Detection (VAD) • High-Fidelity Audio Buffer

Empirical Evaluation & Benchmarks

Law Enforcement Precision & Latency Benchmarks

Hackathon Evaluation
Winner PS2

Selected as the Winner of Problem Statement 2 (Law Enforcement Transcription) in the BHASHINI LEAP Hackathon.

Gemma 3 Slang Fix
94.6% Accuracy

Precision in correcting colloquial slang, dialect idioms, and acoustic mishearings into formal legal phrasing.

BNS Section Precision
96.8% Precision@3

Top-3 penal section recommendation accuracy verified against senior police officer gold test cases.

Draft FIR Latency
74.2 Seconds

Average end-to-end execution latency from completion of spoken interview to fully formatted Form No. 1.

Live Tactical Sandbox

Test the Voice-to-FIR & BNS Recommendation System

Interactive System Workflow Simulator
Live Prototype Demonstration
WINNER: BHASHINI LEAP HACKATHON PS2

Spoken Complaint to BNS 2023 FIR Simulator (Record → Verify → File)

Spoken Hindi / Bhojpuri NarrativeDuration: 00:48
“सर, कल रात करीब 9:30 बजे जब मैं बाजार से लौट रहा था, तब स्टेशन रोड के पास दो अज्ञात लड़कों ने मेरी बाइक रोकी। एक ने कट्टा दिखाया और मेरी पल्सर बाइक और ₹15,000 नकद छीनकर भाग गए।”
Fine-Tuned Gemma 3 Mishearing Correction Engine

Raw ASR detected: “पल्सर पाइप” → Corrected to: “पल्सर बाइक (Motorcycle)” based on police vehicle vocabulary graph.

Structured FIR Draft (<90 Seconds)CCTNS Schema Ready
Incident Type: सशस्त्र डकैती / लूट (Armed Robbery)
Date & Time: Yesterday, 21:30 hrs
Place: Station Road near Main Market
Stolen Property: Bajaj Pulsar Motorcycle, ₹15,000 Cash
Weapon Used: Country-made firearm (कट्टा)
RAG Suggested Sections (BNS 2023):
• BNS Section 309(4): Robbery committed by two or more persons (Earlier IPC 392)
• BNS Section 311: Robbery with attempt to cause grievous hurt or with deadly weapon
Investigating Officer Verification: PENDING REVIEW
Constitutional & Evidentiary Discretion

AI Never Files. The Investigating Officer Decides.

In criminal jurisprudence, an automated allegation without officer review can violate fundamental constitutional rights. MultiLipi bakes mandatory human authority into the software lifecycle:

100% Officer Discretion:The IO can accept, modify, or strike out any suggested BNS penal section.
Digital Signature Sign-Off:No report is transmitted to CCTNS or magistrates without officer token authentication.
Air-Gapped Thana Server:Zero audio recordings or complainant identities ever leak outside the police intranet.
Law Enforcement & Police Modernization

Pilot Voice-to-FIR & BNS Automation in Your Police Circle

Whether you represent a State Police Headquarters, Criminal Investigation Department (CID), Cyber Crime wing, or Police Academy, our team provides on-premise pilot deployments compatible with existing CCTNS infrastructure.