AI / LLM
Voice agents, RAG and multi-step LLM apps—from prompt and tools through speech.
OpenAIGeminiLangChainLangGraphLlamaIndexCrewAIRAGFunction CallingPrompt EngineeringAI AgentsVoice AISarvam AIDeepgramElevenLabsGroq
AI Engineer · Voice & LLM
AayushMalviya.
Mohali, Punjab
I build production voice agents, RAG and LLM products for 10,000+ users—speech pipelines that stay in the 100–150 ms range.
~4
Years building AI
10k+
Users on live products
4+
Products in production
8–10
Developers mentored
Selected work / 2023—2026
Products with a job to do.
TalkEngly, Vokari, Buddy and Massaed—voice practice, support agents, meeting intelligence and elevator ops.
03 / AI conversation intelligence
Buddy
Listens to meetings and turns raw audio into notes, action items, priorities, tags and confidence scores.
Cross-meeting memory in MongoDB so later answers stay consistent with what was already decided.
PythonDeepgramSarvam AIMongoDB
Skills & stack
Voice, retrieval, and the system around them.
LLMs and speech up front, then the Python and Node services, data stores and AWS that keep them in production.
Backend
Python and Node services, real-time channels and queues that keep the request path light.
PythonFastAPINode.jsExpress.jsTypeScriptWebSocketsBullMQREST APIs
Frontend
Product UI on the web, plus desktop and mobile shells when the product needs them.
ReactNext.jsTailwind CSSElectronCapacitor
Vector DB & Databases
Retrieval stores next to the product data the agents actually remember.
PineconeQdrantMongoDBMySQLRedis
Cloud & DevOps
AWS, containers and CI so a voice or LLM feature can stay up after it ships.
EC2S3SESLambdaECSCloudFrontDockerNginxGitHub ActionsGitLab CI/CD
Integrations
Calling, payments and identity wired into the same products.
TwilioTelnyxStripeRazorpayFirebase AuthJWTOAuth
Experience
From first commit to production.
About 4 years owning voice agents, RAG and LLM products from architecture through cloud deploy.
AI Full-Stack Developer
Psquare Company
Mohali, Punjab
FEB 2024 — PRESENT
Own architecture, build and deploy across 4+ products, and mentor 8–10 developers. Massaed’s permission-aware analytics chatbot cut report lookup from 15 minutes to under a minute. Redis and BullMQ moved heavy jobs off the request path and improved API times by 40%. Tathasthu runs LangChain image pipelines on FastAPI. Tudoor and the Electron app Suryacon shipped from scratch, with CI/CD on AWS, Docker and Nginx—about 70% less manual deploy work and 99.9% uptime.
Tudoor — cross-platform app, scratch to productionMassaed — AI analytics chatbot and microservicesTathasthu — FastAPI and LangChain image pipelines
Suryacon — Electron construction management, scratch to production
AI Full-Stack Developer
UrbanChat.ai
Remote
FEB 2023 — JAN 2024
Production voice calling agents for 10,000+ users—real-time STT/TTS, recording and live transfers. RAG on Pinecone, plus function calling, prompt work and sentiment analysis. WebSockets in front of Telnyx, Twilio, Calendly, Cal.com and Stripe.
Sarvam AIElevenLabsDeepgramPineconeTelnyxTwilioStripe
Education
B.Tech · Computer Science
Rajiv Gandhi Proudyogiki Vishwavidyalaya (RGPV)
BHOPAL
CGPA 9.0 / 10.0