Skip to content
Deepak PriyadarshiDeepak Priyadarshi
Ventures
Live

Cogneora

AI agents that answer your calls, chat with visitors and follow up for you.

My role
Founder · built solo
Since
2025
Built for
Clinics and hospitals, real estate, coaching institutes and local service businesses.

Overview

Cogneora gives small businesses in India's tier-2 and tier-3 cities an AI team that works in the languages their customers speak. A business uploads its documents and website once; that shared knowledge base powers voice agents that pick up real phone calls in Hindi, English and Hinglish, website chat that cites its sources and captures leads, automated follow-ups, and an AI-SEO tracker that checks whether ChatGPT and other assistants recommend the business.

SaaS priced in rupees: Starter ₹999, Pro ₹1,999 and Business ₹9,999 a month, with usage-based AI credits.

What it does

  • AI receptionist and telecaller on real Indian phone numbers: answers, books appointments, transfers, records, transcribes and summarises
  • Website chat agent with cited answers and lead capture, embeddable with one script tag
  • No-code workflow builder for follow-ups
  • AI SEO / GEO: weekly visibility score across AI assistants, with fixes
  • Content studio: writers plus image, video and voice generators
  • Social media manager and a public Calls API with webhooks and OpenAPI 3.1

Under the hood

  • Real-time voice pipeline on Pipecat: Sarvam speech-to-text → Groq LLM → Sarvam text-to-speech, with voice-activity detection so callers can interrupt
  • Calls arrive over browser WebRTC or Exotel phone lines over WebSocket at 8 kHz
  • Production RAG without LangChain: multilingual e5 embeddings running in-process via ONNX, hybrid pgvector HNSW + full-text search merged with reciprocal rank fusion, cited answers, SSRF-safe crawling
  • Cost-aware model router with a main and backup model per feature and automatic fallback
  • Usage metering in rupees: credits are reserved per AI call and settled to actual cost; GST invoices generated as PDFs
  • Outbound calling with retries, do-not-call list, QA scoring and lead scoring
  • ~83k lines across API, voice service, webapp and admin, built in about six weeks