AI for Municipal and Swachh Grievance Helplines: Fixing India's Urban Complaint Gap

Executive Summary

India's urban local bodies — from municipal corporations in Maharashtra to city development authorities in Andhra Pradesh — collectively receive hundreds of thousands of citizen complaints every year through a patchwork of helplines, WhatsApp bots, and mobile apps. The complaint categories are structurally simple: garbage not collected, pothole unreported, street light out, water supply disrupted, Swachh Bharat bin overflowing. Every one of these queries has a defined resolution pathway. Yet the median municipal helpline in India resolves fewer than half its complaints within the promised timeframe, and in some cities citizens have simply stopped calling because they expect nothing to happen.

Executive Callout Mumbai's BMC 1916 helpline processes approximately 4,000 calls daily with a ratio exceeding 500 complaints per operator — a workload that produces 30-day average resolution times for a 20-million-person city. Ahmedabad Municipal Corporation has issued six concurrent AI tenders through SCADL, making it India's most procurement-active urban body for AI citizen services. Chennai's Greater Chennai Corporation 1913 helpline runs on 10 operators for 400+ daily complaints, with citizens telling local media they have "stopped calling" because nothing gets resolved. Voice AI handling the five highest-frequency municipal complaint categories — garbage, potholes, water, street lights, and drainage — in local languages and dialects can absorb 60–70% of this call volume without a human agent, enabling the remaining operators to focus on complex escalations. The technology is procurement-ready; the barriers are institutional, not technical. (Aisewak Government Helpline Report, 2026, citing BMC documentation, SCADL tender portal, Bonton PGRS data.)

Introduction

A resident of Dharavi files a complaint about an overflowing garbage bin on Monday. She calls BMC 1916, waits through a queue, reaches an operator, and gives her address. The operator logs the complaint. A reference number is generated. Thirty days later — the standard resolution SLA — the bin may or may not have been emptied. She has no way to check status without calling again, waiting through the queue again, and finding an operator willing to look up her case number.

This is not a technology problem. It is a design problem — and it is one that Voice AI can fix without a multi-year procurement cycle or a greenfield infrastructure investment.

Municipal helplines are structurally different from emergency services like 108 or complex grievance systems like CPGRAMS. The complaint categories are narrow and well-defined. The resolution pathways are known. The backend data — ward boundaries, field team assignments, sanitation crew schedules, water supply zones — exists in municipal ERP systems. The missing piece is a voice interface that connects a citizen in her own language to that backend, at any hour, without a human intermediary.

This article examines the scale of India's municipal helpline failure, the AI opportunity it creates, and what a credible implementation roadmap looks like for Municipal Commissioners and Smart City CEOs evaluating AI for citizen services.

Current Challenges

The Three-City Failure Pattern

Mumbai, Ahmedabad, and Chennai illustrate three distinct failure modes that recur across India's urban local bodies.

Mumbai BMC 1916: Volume without resolution. Mumbai's Brihanmumbai Municipal Corporation operates the 1916 helpline for a population of 20 million, processing approximately 4,000 calls daily (Aisewak Government Helpline Report, 2026, citing BMC data and Mid-Day coverage). With operator ratios exceeding 500 complaints per person per day, the mathematical impossibility of timely resolution is structural, not performance-related. The average resolution time is 30 days — unacceptable for complaints like an open drain during monsoon or a street light outage in a residential lane. In April 2026, BMC launched MARG (Multi-Access Redressal for Grievances), a seven-channel integration system, acknowledging publicly that the single-channel 1916 helpline was failing. Yet MARG does not solve the operator bottleneck — it routes more complaints to the same constrained human workforce.

Ahmedabad AMC 155303: Procurement ambition without voice capability. Ahmedabad Municipal Corporation, through the Smart City Ahmedabad Development Limited (SCADL), has issued six concurrent AI tenders — more than any other urban local body in India (Aisewak Government Helpline Report, 2026, citing AMC tender portal). The Comprehensive Complaint Redressal System (CCRS) RFP, with an EMD of Rs 25 lakh, explicitly requires AI analytics and GIS integration. The 155303 helpline processes 1,500–2,000 daily calls. What is absent from Ahmedabad's ambitious procurement stack is a voice-native interface: a citizen calling 155303 still reaches a human operator queue, not a conversational AI system that can register complaints in Gujarati, auto-categorise by ward, and provide real-time status updates. The six AI tenders demonstrate appetite; the missing voice layer is the deployment gap.

Chennai GCC 1913: Abandonment. The Greater Chennai Corporation's 1913 helpline is a case study in demand destruction. With only 10 operators handling 400+ daily complaints, the queue times have driven citizens to stop calling entirely — a pattern documented in local media and Bonton PGRS analysis (Aisewak Government Helpline Report, 2026). An ongoing revamp will expand to 90 operators, but this human-scaling approach requires sustained hiring, training, and retention investment to serve a city of 11 million — an approach that addresses symptoms without solving the underlying structural mismatch between complaint volume and resolution capacity.

The Swachh Bharat Accountability Gap

The Swachh Bharat Mission — launched in 2014 with the objective of eliminating open defecation and improving solid waste management across India — created a national mandate for urban cleanliness accountability. Urban local bodies receive daily citizen reports on garbage collection failures, open defecation sites, drainage blockages, and public toilet maintenance through Swachh Bharat portals and local helplines.

The Swachh complaint mechanism has two structural weaknesses. First, digital-only submission excludes the majority of urban poor — migrant workers, domestic staff, elderly residents — who interact most directly with the cleanliness infrastructure but are least likely to navigate a mobile app. Second, complaint status visibility requires portal login credentials that most citizens do not maintain. Voice AI addresses both weaknesses simultaneously: a toll-free call in the citizen's language, with Aadhaar or mobile number for identity, and an automated callback for resolution confirmation, is accessible to the full urban population regardless of digital literacy.

Why Traditional Municipal Helplines Fail

The failure pattern across municipal helplines follows four consistent dimensions, documented across why government helplines fail:

Narrow staffing for broad mandates. Municipal corporations serve populations of 2 million to 20 million through contact centres sized for 10–500 operators. No private-sector customer service operation would staff a 20-million customer base with 10 agents. The institutional assumption that citizen complaints are lower priority than commercial customer service has produced a chronic under-investment in urban helpline capacity.

Language exclusion. Municipal corporations in linguistically diverse cities — Mumbai (Marathi, Hindi, Gujarati, Urdu), Chennai (Tamil, Telugu, Malayalam), Ahmedabad (Gujarati, Marwari, Hindi) — predominantly operate helplines in Hindi or English. Citizens whose primary language is Marwari in Ahmedabad's Walled City or Tamil in Chennai's migrant-worker communities face a language barrier at the first point of contact. The multilingual Voice AI for Bharat article details how Bhashini's production infrastructure makes city-level language coverage achievable without building separate systems per language.

No resolution verification. As documented in the AI vs traditional government call centres comparison, municipal helplines uniformly measure complaint disposal (marked "closed" in the system) rather than resolution (citizen confirms the problem is fixed). A garbage bin marked "complaint attended" by a field crew that drove past without emptying it enters the system as resolved. An automated AI callback — calling the citizen 48 hours after the field crew's action — creates the verification loop that transforms disposal into resolution.

Query concentration without AI routing. Five complaint categories — garbage collection, pothole, street light, water supply, and drainage — account for the majority of municipal helpline volume across every city. These categories have defined resolution pathways and backend data systems. Yet municipal IVRs route all complaints to a human queue rather than resolving high-frequency, structured queries at point of contact.

How Voice AI Solves the Municipal Grievance Problem

The Five-Layer Municipal Voice AI Architecture

A production-grade municipal voice AI system operates across five integrated layers tailored to urban local body workflows:

Layer 1: Voice-First Multilingual Intake. The citizen calls the municipal helpline. The AI identifies language preference and greets in Gujarati, Marathi, Tamil, or the citizen's chosen dialect. No menu navigation required — the citizen simply describes the problem in their own words. Natural language understanding categorises the complaint automatically.

Layer 2: Geo-Location and Ward Mapping. The citizen provides their address or confirms a GPS-pinged location if calling from a smartphone. The system maps the address to the correct ward, field crew zone, and responsible officer — a step that currently requires a human operator to consult ward maps manually, creating both delay and error risk.

Layer 3: Backend Integration and Immediate Logging. The complaint is logged directly into the municipal ERP or complaint management system (Bonton PGRS, ServiceNow, NIC state platform, or the city's custom system) in real time. The citizen receives a reference number by voice and SMS. No human agent is required for this step on standard complaint categories.

Layer 4: Field Crew Assignment and SLA Tracking. The system assigns the complaint to the relevant field crew based on workload distribution and confirms the expected resolution time to the citizen. SLA countdown begins from registration, not from when a human operator eventually opens the queue.

Layer 5: Automated Resolution Verification. After the field crew marks the complaint as attended, the AI calls the citizen to confirm resolution: "Your drainage complaint at [address] was attended on [date]. Has the problem been resolved? Press 1 for yes, press 2 to reopen." Non-verified closures are automatically escalated.

Seasonal and Event-Driven Demand Spikes

Municipal helplines face predictable, calendar-driven demand surges that expose the structural limits of human-only contact centres. Garbage complaints spike during Diwali and Holi — when waste volume is 3–5x the daily average. Drainage and pothole complaints surge during monsoon (June–September). Street light complaints cluster in October–November as evenings shorten.

Voice AI's elastic capacity — handling any call volume without incremental staffing cost — converts these predictable surges from service failures into manageable operational events. A DISCOM contact centre pattern applies here: as documented in AI for DISCOM electricity complaint helplines, the seasonal predictability of municipal demand makes ROI calculations unusually credible, because the comparison between AI and human costs is provable within a single surge cycle.

Real Government Use Cases

Ahmedabad: Six Tenders, One Missing Layer

Ahmedabad's SCADL procurement track represents the most analytically useful case because it demonstrates the gap between AI ambition and voice capability. The six concurrent AI tenders include analytics, GIS integration, app-based tracking, and citizen sentiment tools — but none specifies a voice-first complaint intake system (Aisewak Government Helpline Report, 2026, citing SCADL/2025-26/03 CCRS RFP).

This gap is strategic for any voice AI vendor. Ahmedabad's CTO Prithvirajsinh Zala has publicly committed to Smart City AI transformation. The procurement infrastructure — vendor pre-qualification, EMD mechanisms, technical specifications — is already operational. Positioning a voice AI layer as the citizen-facing front-end of the CCRS backend aligns with Ahmedabad's stated programme objectives without requiring a new procurement category.

The replication value extends beyond Ahmedabad: Gujarat has 8 municipal corporations and 162 municipalities, most of which lack any AI-enabled citizen service capability (Aisewak Government Helpline Report, 2026). A reference deployment in Ahmedabad creates a template procurement document for the entire state.

BMC Mumbai: The MARG Integration Opportunity

BMC's April 2026 MARG launch — integrating seven citizen complaint channels — created an integration architecture that voice AI can connect to without a new infrastructure decision. The Municipal Commissioner, Smt. Ashwini Bhide, had already floated a Generative AI tender in 2024, demonstrating that voice and AI are within the institutional procurement mandate (Aisewak Government Helpline Report, 2026, citing BMC data).

The AI opportunity is not to replace MARG but to add a voice-native intake layer: a citizen calls the MARG number, speaks in Marathi or Hindi, and the voice AI registers the complaint in MARG's backend with ward routing and SLA assignment. The voice channel serves the 60–70% of Mumbai's urban poor who are most affected by municipal service failures but least likely to use a smartphone app or WhatsApp bot.

International Benchmark: Singapore's OneService

Singapore's OneService platform — a government-developed municipal complaint system launched in 2015 — processed 500,000+ cases by 2022 with an average resolution time of under 7 days for standard categories (Singapore Ministry of National Development). The key driver of its efficiency is not AI: it is structured categorisation and backend routing to the correct agency in real time. Voice AI applied to India's municipal context adds a language layer on top of this structural approach — addressing the digital literacy barrier that makes Singapore's app-first model inapplicable to India's urban population mix.

Implementation Roadmap

A Municipal Commissioner evaluating voice AI for their helpline can follow a four-phase approach that generates demonstrable ROI within a single budget cycle:

Phase	Timeline	Action	Success Metric
Pilot Design	Weeks 1–4	Define 5 complaint categories; integrate voice AI with existing ERP; configure ward routing	System integration complete; test calls passing
Live Pilot	Weeks 5–10	Deploy on one ward or zone; handle inbound complaints in target language	500+ calls handled; first-call registration rate >90%
Measurement	Weeks 11–12	Compare resolution time, callback confirmation rate, and citizen satisfaction vs. baseline	Resolution time reduction; satisfaction >65%
Scale	Month 4–12	Roll out city-wide; add languages and complaint categories	Full helpline integration; operator load reduced 50%+

The pilot design phase is critical. Municipal ERPs vary significantly by city — some use NIC-built state platforms, some use Bonton PGRS, some use custom systems. Voice AI integration requires API access to the complaint management backend. Confirming API availability before the pilot begins prevents the most common implementation delay.

Expected Impact and ROI

The ROI case for municipal voice AI rests on three calculable dimensions:

Operator cost reduction. At a fully-loaded cost of Rs 25,000–40,000 per month per operator (including training, supervision, attrition, and infrastructure), a helpline handling 4,000 daily calls with 60% AI containment frees approximately 10–15 operator-equivalents. At Rs 30,000 per operator-month, this is Rs 3.6–5.4 crore annually — against an AI voice platform cost of Rs 50 lakh to Rs 2 crore per year at standard government pricing of Rs 2–5 per call (Aisewak Government Helpline Report, 2026).

Resolution time compression. AI complaint intake with automated SLA tracking reduces the time between complaint registration and field crew assignment from 24–72 hours (typical for human-operated queues) to under 30 minutes. For time-sensitive complaints — flooding during monsoon, open manhole after dark — this is a measurable public safety improvement, not only a service quality metric.

Citizen satisfaction and political visibility. Municipalities that deploy AI citizen service solutions receive positive media coverage and serve as model cities in national rankings — Smart Cities Mission assessment, Swachh Survekshan, and the Ease of Living Index all include citizen service responsiveness as evaluation criteria. A Municipal Commissioner who can demonstrate reduced complaint resolution time in Swachh Survekshan data is making a career-advancing decision, not only an operational one.

Risks and Mitigation

Risk	Likelihood	Mitigation
ERP integration complexity delays deployment	Medium	Confirm API access in week 1; scope pilot to categories with clean data flows first
Language coverage gaps in dialects	Medium	Use Bhashini APIs for scheduled languages; add dialect training data for city-specific variants
Field crew non-compliance with status updates	High	Mandate ERP closure with photo evidence; AI callback triggers if crew marks closure without confirmation
Political sensitivity around complaint volume visibility	Low	Frame AI dashboards as operational transparency, not surveillance; Municipal Commissioner controls data access
Citizen trust in AI for grievance	Low	Keep human escalation pathway visible; AI introduces itself as a government automated assistant

Future Outlook

India's Smart Cities Mission and AMRUT 2.0 are deploying Integrated Command and Control Centres (ICCCs) across 100 cities. These ICCCs aggregate real-time urban data — traffic, utilities, weather, citizen complaints — into a single operational dashboard. Voice AI is the citizen-facing input layer that feeds this infrastructure: every call converted to structured, geo-tagged, categorised data in real time.

As ICCCs mature, the distinction between a complaint helpline and a real-time urban management system will dissolve. A citizen reporting a road cave-in by voice will trigger not only a complaint ticket but an automatic alert to the roads department field control room and a traffic management rerouting instruction — all from a single phone call. The governance model for this integrated future depends on voice AI being deployed and refined now, during the window when individual helplines are the unit of procurement.

The governance AI maturity model frames this trajectory as a three-stage progression: reactive helplines (current state for most cities), predictive citizen service (AI initiating contact before complaints arise), and integrated urban intelligence (voice as the primary citizen data input to city management). Municipal corporations that invest in voice AI for helplines today are building the foundation for the second and third stages, not only optimising the first.

Key Takeaways

Municipal helplines in India's largest cities operate with operator-to-complaint ratios that make timely resolution structurally impossible without AI intervention.
The five highest-frequency municipal complaint categories — garbage, potholes, street lights, water supply, drainage — are structurally suited to voice AI automation, with defined resolution pathways and backend data systems already in place.
Ahmedabad's SCADL has issued six concurrent AI tenders; BMC Mumbai launched MARG with Generative AI ambitions; both represent near-term deployment opportunities for voice AI as the citizen-facing layer of existing AI investments.
Resolution time, not disposal rate, is the correct KPI for municipal AI — and automated callback verification is the mechanism that closes the gap between bureaucratic closure and citizen confirmation.
Smart Cities Mission ICCCs and AMRUT 2.0 create the data infrastructure for integrated urban voice AI; municipal helpline deployments today establish the reference architecture for this integrated future.

Conclusion

India's urban local bodies face a citizen service crisis driven by a structural mismatch: complaint volumes growing with urbanisation, budgets constrained by debt and competing priorities, and human contact centres that cannot scale to meet the gap. Voice AI for municipal helplines is not a pilot-stage technology — it is a production-ready capability that Ahmedabad, Mumbai, and Tamil Nadu Smart City programmes are already procuring, if not yet in voice-native form.

The Municipal Commissioner or Smart City CEO who deploys a voice AI layer on their grievance helpline before the next Swachh Survekshan assessment cycle gains three compounding advantages: measurable reduction in complaint resolution time, citizen satisfaction data that supports national ranking performance, and a replicable procurement template that positions the city as a governance innovator in the national Smart Cities programme.

Government leaders exploring AI-powered citizen engagement can begin with a focused pilot in one department or constituency to validate impact before scaling statewide. Aisewak helps public institutions deploy multilingual Voice AI solutions designed specifically for Indian governance.

FAQ

Q1: Which municipal helpline numbers are most commonly used in India? The most prominent municipal helplines include BMC Mumbai's 1916, Ahmedabad's 155303, Chennai GCC's 1913, Pune Municipal Corporation's helpline (020-25503399), and the national Swachh Bharat grievance portal. Most state capitals also operate city-specific grievance numbers through their urban local bodies.

Q2: What complaint categories are most suitable for Voice AI automation in municipal helplines? Garbage collection failures, pothole reports, street light outages, water supply disruptions, and drainage blockages collectively account for the majority of municipal helpline volume across Indian cities. All five have defined resolution pathways and backend data systems, making them structurally suited to Voice AI automation without human agent involvement for first-contact handling.

Q3: How does Voice AI handle complaints from citizens who speak regional languages or dialects? Voice AI systems built on Bhashini's multilingual infrastructure — which supports 22 Indian languages in voice production — can handle complaints in Marathi, Tamil, Gujarati, Telugu, Kannada, and other scheduled languages. City-specific dialect variants (Marwari in Ahmedabad, Bhojpuri in Varanasi) require additional training data but are achievable with 4–6 weeks of localisation work.

Q4: What is the typical ROI timeline for a municipal Voice AI helpline deployment? A pilot covering one zone or ward with 500+ daily calls can demonstrate measurable ROI within 60–90 days: operator load reduction, complaint registration speed, and resolution time verification are all quantifiable within a single operating cycle. Full citywide ROI — including platform costs amortised against operator savings — typically turns positive within 12–18 months at standard government pricing of Rs 2–5 per call.

Q5: How does Voice AI integrate with existing municipal ERP and complaint management systems? Voice AI platforms integrate with municipal backends via REST APIs. Common target systems include Bonton PGRS (used by several large corporations), NIC state grievance platforms, ServiceNow, and city-specific custom systems. API availability confirmation is the critical dependency in the pilot scoping phase — most modern municipal ERPs support API access; legacy systems may require a data bridge.

Q6: Can Voice AI handle the Swachh Bharat Mission grievance submission process? Yes. Citizens can report Swachh Bharat-relevant complaints — open defecation sites, overflowing bins, public toilet maintenance — through a voice interface that logs directly to the Swachh Bharat Mission portal or the city's PBMS (Performance-Based Management System). The voice channel extends Swachh Bharat accountability to citizens who cannot navigate the Swachh app, including elderly residents and migrant workers.

Q7: What procurement route should a Municipal Commissioner follow to deploy Voice AI? Municipal corporations can procure Voice AI through Smart City Mission ICCC RFPs (for cities in the Smart Cities programme), AMRUT 2.0 technology grants, direct competitive tender under GFR, or through state government NICSI channel work orders. Ahmedabad's SCADL procurement structure — concurrent AI tenders with defined technical specifications — is a replicable model for other cities with Smart City Special Purpose Vehicles.

Q8: How does Voice AI improve Swachh Survekshan and Ease of Living Index scores? Both national assessments include citizen complaint resolution time and satisfaction as evaluation criteria. AI-enabled automated callback verification creates a documented satisfaction data trail — distinct from disposal-rate claims — that can be submitted as evidence in Swachh Survekshan questionnaire responses. Cities with demonstrably reduced resolution times and higher citizen satisfaction rates in documented surveys typically receive higher scores on citizen service sub-indices.

Schema Markup Suggestions

Article (schema.org/Article): Primary type for the blog post. Key fields: headline, description, author, datePublished, dateModified, publisher.
FAQPage (schema.org/FAQPage): Apply to the FAQ section. Each Q&A pair maps to a Question + Answer entity.
GovernmentService (schema.org/GovernmentService): Applicable to references to BMC 1916, Ahmedabad 155303, Chennai GCC 1913, and Swachh Bharat grievance mechanisms.
GovernmentOrganization (schema.org/GovernmentOrganization): For entities including BMC, AMC, GCC Chennai, SCADL, Smart Cities Mission, AMRUT programme.
HowTo (schema.org/HowTo): Applicable to the Implementation Roadmap section, structured as a four-phase how-to guide.

Suggested External References

BMC MARG launch documentation (April 2026), Brihanmumbai Municipal Corporation
SCADL/2025-26/03 CCRS RFP, Smart City Ahmedabad Development Limited
Swachh Survekshan assessment framework, Ministry of Housing and Urban Affairs
AMRUT 2.0 technology grant guidelines, Ministry of Housing and Urban Affairs
Smart Cities Mission ICCC implementation guidelines, MoHUA
Singapore OneService platform, Ministry of National Development (annual performance review 2022)
Bonton PGRS grievance management documentation
Aisewak Government Helpline Report, 2026 (internal research, 300+ searches, CAG, RTI, parliamentary questions)

X / LinkedIn caption: India's municipal helplines handle hundreds of thousands of garbage, pothole, and drainage complaints — with operator ratios so high that resolution in 30 days is the best case. Voice AI for civic grievances isn't futuristic: Ahmedabad has 6 AI tenders open, BMC launched MARG, and Tamil Nadu is procuring Smart City ICCC voice layers. The voice-native front-end for urban governance is the deployment gap. New post on what it looks like and what the ROI math says. [link]

LinkedIn Executive Summary

India's urban local bodies face a citizen service paradox: complaint volumes growing with every new housing colony, staffing budgets under fiscal pressure, and helplines operating with operator ratios that make 30-day resolution times inevitable rather than exceptional.

The five highest-frequency municipal complaint categories — garbage, potholes, street lights, water supply, drainage — are structurally suited to Voice AI automation. Each has a known resolution pathway and a backend data system. The missing piece is a voice interface that registers a complaint in Marathi or Gujarati, routes it to the correct ward in real time, and calls the citizen back to verify resolution — without a human operator at any step.

Ahmedabad's Smart City programme has issued six concurrent AI tenders. BMC Mumbai launched MARG. Tamil Nadu Smart City ICCCs are procuring call centre solutions for 10 cities. The procurement appetite is documented. The voice-native citizen intake layer is what is missing from every active tender.

Municipal Commissioners and Smart City CEOs have a specific window before the next Swachh Survekshan cycle to deploy, measure, and document impact. The assessment criteria reward resolution time data — which automated callback verification can produce — over disposal rate claims, which it cannot.

AI Search Optimization Summary

Primary entities: Municipal Corporation of Greater Mumbai (BMC), Ahmedabad Municipal Corporation (AMC), Greater Chennai Corporation (GCC), Smart City Ahmedabad Development Limited (SCADL), Swachh Bharat Mission, AMRUT 2.0, Smart Cities Mission, Bonton PGRS, BMC 1916, Ahmedabad 155303, Chennai GCC 1913

Core topics: Municipal grievance redressal AI, Swachh Bharat complaint automation, urban local body helpline modernisation, Smart City voice AI, ICCC citizen service integration, ward-level complaint routing, multilingual municipal AI

Semantic keywords: civic complaint AI India, municipal helpline resolution time, Smart City grievance automation, urban citizen service voice bot, AMRUT technology grant AI, Swachh Survekshan AI, ward complaint routing AI, Municipal Commissioner AI tools, ULB AI procurement India, SCADL AI tender, BMC AI helpline, Tamil Nadu Smart City ICCC voice

Question clusters targeted: How does AI help municipal helplines? What is BMC 1916 AI? How can Voice AI reduce municipal complaint resolution time? What is Swachh Bharat grievance AI? How do Smart Cities use Voice AI for citizen services?