Sign in securely
Use your assigned account. Sessions last six hours; never share credentials.
Opening page…
MPSEDC/COE/2026/703
Operating procedures, complaint lifecycles, platform modules, controls, and RFP demonstration guidance.
This is an operating and live-demonstration guide. “Demonstrate”, “Verify”, and “Evidence required” labels describe readiness; they do not certify qualification by the Evaluation Committee.
Quick start
Use your assigned account. Sessions last six hours; never share credentials.
Open Dashboard and confirm API, connector, wallet, telephony, and agent readiness.
Use consent, verified tools, official ticket IDs, audit records, and role-based access.
System map
Channels
Exotel · SIP · WebRTC · WhatsApp
Voice runtime
STT → Flow/LLM/RAG/Tools → TTS
Government systems
CM Helpline · eSEVA · SMS · Databases
JWT/RBAC, consent, PII redaction and audit are present; verify guardrail enforcement on every runtime path.
Hindi-first, multilingual agent settings, Indian-language ASR/TTS paths.
CDRs, recordings, transcripts, tool runs, dispositions, reports, versions.
Citizen → ticket → resolution
Goal: collect the minimum details and consent, register through the connected grievance system, provide its confirmed ticket, and track the case through resolution.
The public number routes through Exotel/SIP to the MPSEDC inbound voice agent.
The agent starts in Hindi and identifies new registration, status enquiry, unresolved complaint, feedback, or human assistance.
Collect citizen name, phone, district, category/department, and a concise description. Ask one question at a time.
Summarise the issue and obtain explicit consent before submission. Consent is mandatory in the grievance API.
Invoke the complaint-registration tool/CM Helpline connector. Do not replace registration with an unsupported refusal.
Read the confirmed tracking ID clearly, repeat it on request, and send acknowledgement where SMS is configured.
Operators retrieve by tracking ID, refresh external status, add verified updates, and escalate according to workflow.
A resolution note is required before resolved/closed status. Retain call, action, and audit evidence.
Partially Closed feedback
Pass name and ticket_id as canonical call variables. Close satisfied calls; for unresolved cases, obtain consent and update the grievance or create a confirmed new ticket.
Use Partially Closed complaints, respect DNC/quiet hours, and provide each contact's name and original ticket ID.
Choose the MPSEDC outbound agent, schedule calling slots, set concurrency, then start, pause, resume, or stop.
The female Shruti voice uses feminine Hindi and receives {{name}} and {{ticket_id}} as call variables.
Ask whether the original issue is resolved and whether the citizen is satisfied. Never infer satisfaction from status alone.
Record SATISFIED, thank the citizen, close respectfully, and retain the disposition for reporting.
Capture the remaining issue, acknowledge it, obtain consent, then update the existing grievance or register a new ticket.
Read only a system-confirmed new ticket/action. Never invent a number, resolution promise, or timeline.
Review transcript, recording, analysis, disposition, campaign report, and schedule governed follow-up when required.
State control
Ticket accepted
Department action
Resolution note required
Final verified closure
Operator console
RBAC
| Role | Responsibility | Rule |
|---|---|---|
| admin | Full platform configuration, users, agents, channels, and governance. | Least required access |
| operator | Calls, grievance intake, citizen assistance, and operational actions. | Least required access |
| auditor | Read-only oversight, analytics, call records, and audit evidence. | Least required access |
| content_approver | Maker-checker review and approval of flows, prompts, and content. | Least required access |
| security | Security settings, PII controls, credentials, and incident oversight. | Least required access |
| campaign_manager | Audience, schedule, concurrency, DNC, and campaign lifecycle. | Least required access |
TQ §4.1
Demonstrate
Live UI/runtime path exists
Verify
Integration or complete proof must be checked
Evidence required
External/signed documentary artifact
Implementation gap
Required runtime capability was not found
| ID | Criterion | Evaluation | Readiness | Demo / evidence path |
|---|---|---|---|---|
| I.1 | Self-service conversational agents with STT, TTS, LLM responses, and voice interactionsI · Self-Service Authoring | Demo + compliance | Demonstrate | Agents, Voice AI, Calls |
| I.2 | State-based design, tool access, and variable access controlsI · Self-Service Authoring | Demo + compliance | Verify | Flows + Tools; verify per-state ACL evidence |
| I.3 | No-code admin panel for creating and managing conversation flowsI · Self-Service Authoring | Demo + compliance | Verify | Builder exists; verify schema-compatible runtime deployment, approval, and rollback |
| I.4 | Custom knowledge base with RAG and API/tool callingI · Self-Service Authoring | Demo + compliance | Verify | Knowledge and tool UIs exist; verify indexed RAG answer with citation on the live runtime |
| I.5 | CRM, SMS, email, government application, and database integrationsI · Self-Service Authoring | Demo + compliance | Verify | Connectors/tools; use production credentials and sample API |
| I.6 | Automatic detection and switching across Indian languagesI · Self-Service Authoring | Demo + compliance | Verify | Language configuration + live multilingual call |
| I.7 | Code-mixed translation with entity preservationI · Self-Service Authoring | Demo + compliance | Gap | No complete translation/entity-preservation runtime found; implement and evidence it |
| I.8 | Regression testing and AI evaluation frameworkI · Self-Service Authoring | Demo + compliance | Verify | Simulation/refinement + signed evaluation output |
| II.1 | ASR with noise cancellation, accents, and regional dialectsII · Conversational Capabilities | Demo + compliance | Verify | Recorded noisy/dialect call set |
| II.2 | Voice gender, modulation, empathy, pitch, pace, and pronunciation controlsII · Conversational Capabilities | Demo + compliance | Verify | Voice AI + Agents; verify every requested control |
| II.3 | Formal and conversational speaking stylesII · Conversational Capabilities | Demo + compliance | Verify | Persona/prompt controls exist; verify distinct live speech styles |
| II.4 | Interruptions and merged-turn handlingII · Conversational Capabilities | Demo + compliance | Verify | Live barge-in/flush media trace |
| II.5 | Sentiment detection, dynamic adaptation, and escalationII · Conversational Capabilities | Demo + compliance | Gap | No sentiment-adaptation runtime found; implement before the live qualification demo |
| II.6 | Conversation memory and personalised engagementII · Conversational Capabilities | Demo + compliance | Verify | Session variables + returning-citizen test |
| II.7 | End-to-end voice latency below one second under loadII · Conversational Capabilities | Evidence + demo | Evidence | Dated load report, p95 traces, and dashboard export |
| II.8 | Prescribed CER and Language Identification performanceII · Conversational Capabilities | Documentary evidence | Evidence | Signed CER/LID report on approved datasets |
| III.1 | IVR and cloud-telephony integrationIII · Agent Configurations | Demo + compliance | Verify | Exotel/WebSocket core exists; SIP control is incomplete and needs a live verified path |
| III.2 | Scheduling, load balancing, auto-scaling, and required concurrent callsIII · Agent Configurations | Evidence + demo | Evidence | Campaign concurrency + HPA + external load test |
| IV.1 | Campaign management, segmentation, and schedulingIV · Orchestration & Monitoring | Demo + compliance | Verify | Campaign UI exists; segmentation and scheduler/runtime linkage need end-to-end proof |
| IV.2 | Disposition-based multi-day campaign orchestrationIV · Orchestration & Monitoring | Demo + compliance | Verify | Retry/callback policy + multi-day workflow audit |
| IV.3 | Testing, simulation, and regression managementIV · Orchestration & Monitoring | Demo + compliance | Verify | Flow simulation/refinement + versioned result set |
| IV.4 | PII maskingIV · Orchestration & Monitoring | Demo + compliance | Verify | Caller masking and log redaction exist; verify all exports/media/runtime paths |
| IV.5 | SIP and WebSocket telephony integrationIV · Orchestration & Monitoring | Demo + compliance | Verify | Media WebSocket trace + configured SIP call |
| IV.6 | Responsible-AI controls and misuse guardrailsIV · Orchestration & Monitoring | Evidence + demo | Verify | Injection test, confidence policy, PII and audit evidence |
| V.1 | Automated agent evaluation against business rules and SOPsV · Reporting & Analytics | Demo + compliance | Gap | No executing SOP evaluator/exportable report found; implement before qualification |
| V.2 | Role-based analytics, audit logs, and reportingV · Reporting & Analytics | Demo + compliance | Verify | RBAC, analytics and audit listing exist; role-filtered UI/export/tamper evidence need proof |
| V.3 | Voice of Citizen and conversational analyticsV · Reporting & Analytics | Demo + compliance | Verify | Analytics trends, outcomes, transcript insights |
Live solution only
Create an agent and flow; add intent, prompt, RAG/tool, fallback, and handoff states; save a draft.
Run Hindi/Hinglish scenarios, then verify the missing regression/SOP and maker-checker gates before claiming this step.
Demonstrate noisy speech, language switch, interruption, DTMF fallback, complaint registration, ticket, and PII masking.
Upload/enter contacts, enforce DNC and slots, start/pause/resume, capture disposition, and create a new ticket when unresolved.
Show CDR, recording, transcript, outcomes, audit trail, prompt-injection guardrail, and evidence export.
Use the supplied API/workflow unchanged, map it to a tool/flow, and exercise it on a live call without a code deployment.
Presentation-only demonstrations are not accepted. The committee may provide an API/workflow; exercise it unchanged on the live solution. Use the same LLM in demonstration and production unless MPSEDC approves an equivalent or superior replacement in writing.
Defence in depth
JWT, six-hour sessions, and backend role checks exist. MFA/OIDC, role-filtered navigation, and maker-checker separation still require implementation or proof.
Obtain consent before grievance submission; collect only required citizen data.
Caller display masking and API-log redaction exist. Redact and authorise exports explicitly; full media/runtime coverage is not yet proven.
PII, injection, and confidence libraries exist, but end-to-end runtime enforcement must be wired and demonstrated.
Request-level audit listing exists. Before/after values, export, immutable storage, and tamper evidence still require completion.
Use MP SDC/data-residency controls, approved egress, secrets, monitoring, HA, DR, and VAPT evidence.
Operator checklist
Failed to fetch
Match NEXT_PUBLIC_API_URL to the running backend port and restart the frontend.
Session expired
Sign in again to receive a new six-hour JWT; check server clock and ACCESS_TOKEN_EXPIRE_MINUTES.
Agent speaks a placeholder
Use {{name}} / {{ticket_id}} in the prompt and send exact call-variable keys.
Wrong Hindi gender
Match grammatical gender to the selected voice (Shruti=feminine, Ashu=masculine).
Ticket not returned
Inspect the complaint tool run and connector response; do not treat an invented ID as external confirmation.
Call not placed
Validate E.164 number, agent publication, DNC status, wallet, schedule, and telephony connection.