Frequently asked questions

We design, build and run production AI systems — not pilots. Every engagement ends with something live in your operations: a phone number an AI answers, a WhatsApp number that qualifies leads, a model scoring your data, or a pipeline taking actions in your CRM.

We build the telephony and infrastructure ourselves rather than reselling a no-code tool. That includes self-hosted Asterisk/SIP where a client needs it, our own multi-tenant WhatsApp Tech Provider platform, and models fine-tuned on client data. It is why our systems handle Indian languages, Indian accents and code-switching properly.

Skanda AI Solutions Private Limited is based at HMT Officers Colony, Old Alwal, Hyderabad, Telangana 500010, India. The company was founded in 2024. CIN U62013TS2026PTC216311, GSTIN 36ABTCS7747M1Z7. We serve clients across India as well as in the United States, United Kingdom, UAE, Australia and Singapore.

Karthik Raj Gupta is Managing Director & CEO. Builds and ships the voice, conversational and applied-ML systems Skanda AI deploys for clients. Shyam Vishwanatham is Chief Operating Officer. Runs delivery and client operations across Skanda AI engagements.

Our rates are deliberately among the most affordable in India, and they are negotiable. Focused voice or WhatsApp deployments typically start around ₹2–3 lakh, mid-complexity agentic systems run ₹5–15 lakh, and larger fine-tuning or computer-vision programmes range ₹15–50 lakh — but those are starting points, not a fixed price list. Our WhatsApp platform is subscription-priced with no per-seat licensing. Tell us your budget and we will tell you honestly what fits inside it. If you have a cheaper quote from another vendor, send it to us and we will work to beat it — we will not lose your project on price.

Usually, yes, and deliberately so. We price for Indian businesses rather than for enterprise procurement: no per-seat licensing on the WhatsApp platform, no charge per feature, and no markup on Meta’s own WhatsApp conversation fees. Custom development is quoted per module rather than as an enterprise change request. If you already have a quote from another vendor, send it to us — we will work to beat it. We would rather win the work at a smaller price than lose it on cost.

Yes. Every AI system we build is configured to your business, and our WhatsApp platform additionally supports client-specific modules that run on top of it — approval workflows, bespoke booking logic, scheduled reminder jobs, custom data extraction and integrations with whatever systems you already run. One live deployment runs an entire dental appointment lifecycle this way. If your process does not fit the standard product, we change the product rather than asking you to change your process.

Voice agents and WhatsApp automations reach production in two to three weeks. Computer vision takes four to eight weeks. Agentic pipelines and LLM fine-tuning take eight to twelve weeks. Our fastest production deployment was eleven days.

Both. Some of our best deployments are with growth-stage startups and family-run businesses — a ten-branch jewellery chain, a dental clinic, a two-wheeler dealership, a Telugu newspaper. Engagements start around ₹2–3 lakh. What matters is having a real operational problem, not company size.

Both. The same team builds the model and the product it lives in — the website, the dashboard, the admin tools, the telephony layer. We have shipped e-commerce platforms, on-premise valuation software, marketing sites and internal tools alongside the AI systems that run inside them.

RAG (Retrieval-Augmented Generation) connects a language model to your live data through a vector database, and is the right choice whenever the underlying facts change — documents, policies, catalogues, internal wikis. Fine-tuning bakes domain language and behaviour into the model weights, and is the right choice for specialised jargon, consistent output format, and cutting inference cost at high volume. Skanda AI recommends RAG for most use cases and fine-tuning where volume or language specificity justifies it. The two combine well.

Yes. Our voice agents run in English, Hindi and Telugu in production, with Tamil, Kannada, Malayalam, Marathi, Gujarati, Bengali and Punjabi supported. Speech recognition is tuned for Indian accents including code-switching such as Hinglish and Tenglish. Generated speech uses native script rather than Latin transliteration, because transliterated Indic text is badly mispronounced, and digit sequences are formatted so phone numbers are read one digit at a time instead of as quantities in lakhs and crores.

We build it. Alongside Plivo and Twilio integrations we deploy self-hosted Asterisk with PJSIP and ARI-driven call control, with SIP endpoints rendered per DID so outbound calls always present the correct caller ID on the correct trunk. That matters for cost at volume, for caller-ID compliance across many numbers, and for clients whose call data cannot leave their infrastructure.

In our dental clinic deployment the system detects that the captured details are incomplete, calls the patient back, and resumes the same conversation carrying forward the answers it already holds — rather than starting the script from the beginning. Drop recovery like this is the difference between a demo and a production system.

Data is encrypted in transit using TLS 1.3 and at rest using AES-256. Models can be deployed entirely inside your own AWS, Azure or GCP environment, or on your own servers, so no data leaves your infrastructure. We design to India’s DPDPA 2023, GDPR for international clients, RBI and SEBI expectations for financial services, and ABDM for healthcare. We never train on client data without explicit written consent.

Yes, and we have done it. A property valuation system we built runs entirely on the client’s own Ubuntu infrastructure with Nginx and systemd, because valuation case data had to stay in-house. Fine-tuned models, vision systems and edge deployments can all run inside your environment.

Skanda AI is an approved Meta WhatsApp Business Tech Provider. We connect your own WhatsApp Business Account to the official WhatsApp Business API through Meta embedded signup, then run an AI agent on that number: answering messages 24/7, qualifying leads, capturing contact details, booking appointments, tracking orders and sending follow-ups in Hindi, Telugu, Tamil or English, with handover to your team when a human is needed. You own the number and the account throughout.

For routine, high-volume calls, yes. Our dealership deployment handles 100% of inbound calls with AI — booking test rides, qualifying leads and answering product questions — transferring only complex cases to staff. The realistic outcome is a 60–100% reduction in human handling of routine calls. For regulated or high-value conversations we more often recommend real-time agent assist, which keeps your person on the call while suggesting the right answer on screen.

Real-time agent assist integrates with your existing dialer. As the customer speaks, their audio streams through our pipeline live and a recommended response appears on the agent’s dashboard, grounded in your approved answers and policies. It raises accuracy and consistency on live calls, shortens the ramp time for new telecallers, and gives quality teams a record of what was recommended against what was actually said — across every call rather than a sampled few.

Far less than most platforms quote. The Skanda AI WhatsApp Platform is priced deliberately at the affordable end of the Indian market, with no per-seat licensing and no charge per feature. WhatsApp conversation charges are paid directly to Meta at their rates, with no markup from us. If you have a cheaper quote elsewhere, send it to us — we will work to beat it. Tell us your budget at karthik@skandaai.in and we will tell you honestly what fits.

Yes — that is specifically who we built for. Clinics, single-branch shops, dealerships, coaching institutes and family businesses run on this platform. There is no enterprise minimum, no per-seat cost, and no requirement to buy a bundle of features you will not use. We would rather work with a small business at a small price than price ourselves out of the market.

Completely. The standard platform gives you the AI agent, inbox, templates, broadcasts, flows and API. On top of that we build client-specific modules — approval workflows, bespoke booking logic, scheduled reminder jobs, custom data extraction, integrations with your existing systems. One live deployment runs an entire dental appointment lifecycle this way: call intake, doctor approval by button tap, automatic confirmation, review capture and follow-up reminders. If your process does not fit the standard product, we change the product.

Your own. Skanda AI is an approved Meta Tech Provider, so we connect your own WhatsApp Business Account through Meta embedded signup. The number, the account and the customer relationship belong to you throughout, and nothing is stranded if you ever leave.

Usually two to three weeks from first conversation to a live AI agent on your own number, including Meta verification and template approval. Simple deployments have gone live faster. The dashboard itself is available immediately at app.skandaai.in once your account is created.

Yes, at any moment. Anyone on your team can take over a conversation from the shared inbox and the AI immediately stands down for that thread until it is released back. Nothing is lost in the handover — your team sees the full history plus an AI summary of what the customer wants.

Yes, through the Developer API. You generate an API key from Settings, and your system posts to one endpoint to send an approved template to any number. Sends are queued, paced and retried for you, and an idempotency reference means a retried call never double-sends. We also build direct integrations with Zoho, Salesforce, Google Sheets, n8n and most ERPs on request.

Live deployments span healthcare and dental clinics, diagnostics, real estate, automotive dealerships, jewellery and retail, and education. The platform is industry-agnostic — the AI agent is configured with your knowledge base, and anything industry-specific is added as a custom module.

For routine, high-volume calls, yes. Our deployments handle appointment booking, lead qualification, product questions and follow-ups end to end, and transfer only genuinely complex cases to staff. One dealership deployment handles 100% of inbound calls with AI. The realistic outcome is a 60–100% reduction in human handling of routine calls, with your team redeployed onto the conversations that actually need a person.

English, Hindi and Telugu in production today, with Tamil, Kannada, Malayalam, Marathi, Gujarati, Bengali and Punjabi supported. Agents detect the caller’s language from their first reply and switch, including natural code-switching such as Hinglish and Tenglish.

Two to three weeks from discovery to a live number in most cases. Our fastest production deployment was eleven days.

An AI voice agent replaces the human on routine calls. Agent assist keeps your human on the call and makes them better at it — the AI listens and suggests, the person decides and speaks. It suits regulated, high-value or emotionally sensitive conversations where you want a human voice but consistent, accurate answers.

Yes — it is designed to integrate with the dialer and agent desktop you already run, rather than asking you to migrate telephony. We tap the live audio stream and render suggestions into the agent’s existing workflow.

That is one of its strongest uses. Live suggestions shorten the learning curve on real calls, and the transcript record of recommended versus actual answers gives trainers concrete, specific coaching material instead of impressions from spot-checks.

We connect your own WhatsApp Business Account to the official WhatsApp Business API through Meta embedded signup, then run an AI agent on that number. It answers messages 24/7, qualifies leads, captures contact details, books appointments, tracks orders and sends follow-ups, in Hindi, Telugu, Tamil or English, and hands over to your team when needed.

Yes, and that is deliberate — you own the number, the account and the customer relationship. As a Meta Tech Provider we connect and manage it for you, but it is never our number or a shared one, so nothing is stranded if you ever leave.

Yes. Captured leads and conversation outcomes can be pushed to Zoho, Salesforce, most ERPs, or any webhook endpoint you provide.

Yes, end to end. Our dental deployment answers the call, captures the appointment, sends the patient a holding message, gets the doctor’s approval on WhatsApp, confirms the patient, asks whether the consultation happened, requests a review and sends follow-up visit reminders — with no staff involvement unless a human is genuinely needed.

The system detects that the captured details are incomplete and calls the patient back, resuming the same conversation with the answers it already holds rather than starting the script again.

Agentic AI plans multi-step tasks, uses tools, calls APIs and corrects itself without human direction. Unlike a chatbot it takes real actions — updating your CRM, booking appointments, sending follow-ups, generating reports. One marketing deployment runs three times the campaign volume at roughly a quarter of the previous manual cost.

The rules are part of the system rather than a training slide. The agent may never raise its voice or shame the debtor, never claim a legal case or credit consequence that has not happened, and never discuss the debt with anyone other than the named person. If someone reports hardship or bereavement the collection script is dropped, and a request to stop calling is acknowledged and recorded rather than argued with.

RAG connects a model to your live data through a vector database and is the right choice for knowledge that changes — documents, policies, catalogues, internal wikis. Fine-tuning bakes domain language and behaviour into the model weights and is the right choice for specialised jargon, consistent formatting, and cutting inference cost at high volume. We recommend RAG for most use cases and fine-tuning where volume or language specificity justifies it. They combine well.

Yes. We fine-tune Llama, Mistral and Gemma on proprietary client data using LoRA and QLoRA, and deploy the result either in our infrastructure or entirely within your own AWS, Azure or GCP environment. Typical motivations are reducing hallucination on domain material, cutting per-token inference cost at volume, and handling Indian-language content that general models handle poorly.

Yes — English, Hindi, Telugu, Tamil, Kannada, Malayalam, Marathi, Gujarati, Bengali and Punjabi. Recognition is tuned for Indian accents including code-switching such as Hinglish and Tenglish, and generated speech uses native script because transliterated Indic text is badly mispronounced by text-to-speech engines.

SEO optimises to rank in a list of links. GEO — Generative Engine Optimisation — optimises to be cited inside an answer generated by ChatGPT, Gemini, Claude or Perplexity. The levers differ: GEO depends on explicit factual statements a model can quote, consistent entity data, comprehensive structured data, an llms.txt index, and allowing AI crawlers rather than blocking them. Most sites do the opposite by default.

Data is encrypted in transit with TLS 1.3 and at rest with AES-256. Models can be deployed entirely inside your own AWS, Azure or GCP environment so no data leaves your infrastructure. We design to India’s DPDPA 2023, GDPR for international clients, RBI and SEBI expectations for financial services, and ABDM for healthcare. We never train on client data without explicit written consent.

Yes. We run AI receptionists for a dental clinic and a diagnostic centre in production today — answering calls, booking appointments, arranging home collection, confirming with the clinician on WhatsApp, and sending review and follow-up reminders afterwards. Deployments are ABDM-compliance ready and can run entirely inside your own infrastructure.

Yes. We built outbound AI sales calling for a commercial developer in Narsingi, Hyderabad — qualifying investors versus business occupiers and booking site visits across 75+ projects and 4M+ sq ft, cutting lead qualification time by 60%. We also run a WhatsApp property assistant for a Hyderabad platform covering buyer search, agent matching, loans and listings.

Yes. We built a dual-channel voice and WhatsApp system for an authorised Hero MotoCorp dealership in Hyderabad. The AI handles 100% of inbound calls — booking test rides, qualifying leads and answering product questions — and delivered 340% more qualified leads within 60 days.

Yes. We run an outbound AI relationship executive for a ten-branch jewellery chain in Hyderabad — calling recent customers for feedback and introducing gold savings schemes in English, Telugu or Hindi, which lifted appointment bookings 60% with return visible in the first week.

We built a 24/7 AI learning agent serving 5,000+ students where satisfaction rose from 3.2 to 4.8 out of 5 and 90% of queries are now resolved instantly. We also build admissions assistants grounded in a curated knowledge file the institute edits directly, which capture leads and route commitment-bearing questions such as exact fees to human counsellors.

It can be, if the rules are enforced by the system rather than left to an agent’s judgement. Our collections agents are built so they cannot shame a debtor, cannot claim a legal case or credit consequence that has not happened, and cannot discuss the debt with a third party. Hardship triggers dropping the script, and a request to stop calling is recorded rather than argued with. Every call leaves a full transcript and audit trail.

Defect detection via computer vision, digital twins for production line simulation, predictive maintenance from equipment telemetry, supply chain demand forecasting, and edge AI deployment that runs on the factory floor without cloud dependency.