Voice AI at scale
Production agents handling 500–600 calls/day on SIP/Asterisk/FreePBX for inbound qualification, outbound follow-ups, live transfer, and CRM handoff.
SIP write-up →AI · voice infrastructure · agent systems
> production voice AI, agent backends, and telephony that ships
I build real-time voice agents, multi-agent architectures, and the SIP/Asterisk infrastructure behind them, not demos. Production systems handling hundreds of calls a day, GH200-scale data pipelines, founding-engineer work at startups, and on-prem AI infrastructure for Andhra Pradesh State Police. ECE at VIT Amaravati; most of my learning happens in production.
The strongest proof points: production scale, leadership, and shipped work.
Production agents handling 500–600 calls/day on SIP/Asterisk/FreePBX for inbound qualification, outbound follow-ups, live transfer, and CRM handoff.
SIP write-up →Building on-prem AI infrastructure for the Andhra Pradesh State Police, model deployment pipelines, training staff on local inference, and agentic workflow loops under data residency constraints. Earlier led backend and on-site tech for AP CID home guard exams.
On-prem write-up →600+ participants, 80-person team. Closed ₹21 lakh sponsorship in under a month, built the website and infra, negotiated the first private VLAN on campus. Inaugurated Recon on 19 May. Two lab exams on the 17th, YC Startup School in Bangalore on the 18th, back on campus to open the fest.
Recon write-up →Selected participant, 17 May: two back-to-back ECE lab exams. 18 May: flew to Bangalore, met founders, discussed products and validation frameworks with the cohort. Afterparty: talked to VCs over drinks. 19 May: flew back and inaugurated Recon 2026.
That week →Founding AI engineer: taste-based recommendation engine, 8+ Pydantic AI agents, customer intelligence (RFM, churn), serving 80+ cafes on FastAPI + PostgreSQL + Redis.
Clink role →Bare-metal GH200 pipeline: Whisper v3 transcription, PII sanitization, multi-agent performance analysis. Pioneered SIP trunking to bridge LiveKit with legacy VICIDIAL.
Indominus Labs →Shipped multimodal AI backend for VISU-X humanoid robot in one month, real-time speech, vision, facial recognition. Presented to CM Chandrababu Naidu at VLaunchPad.
VISU-X repo →Technical talks on agent architectures and LLM security to 100+ attendees per session (Null Vijayawada, NIT Warangal, SRM). Winner HackSRM 2024. 3rd at Post-a-thon, showcased IntelliPost to Union Minister Shri Chandra Sekhar Pemmasani. Led Pixy paper reading sessions.
Full list →Recognition, leadership, and public impact. The full list from my resume.
Building on-prem AI infrastructure for state police operations, local model deployment pipelines, training staff on inference and monitoring, and designing agentic workflow loops with human-in-the-loop guardrails. Previously led backend and on-site infrastructure for AP CID home guard exam systems.
Built and shipped production AI voice agents integrated with SIP/Asterisk/FreePBX telephony stacks for inbound lead qualification and outbound follow-ups. Designed end-to-end call workflows spanning voice → intent detection → CRM actions → calendar booking with GoHighLevel and n8n automations.
Pioneered SIP trunking integration to bridge LiveKit with legacy telephony providers (VICIDIAL). Engineered a scalable data pipeline on bare-metal GH200 to process 1M+ customer call logs with Whisper v3 transcription, PII sanitization, and multi-agent performance analysis.
Developed a custom turn-detection model for Hindi, overcoming a core LiveKit SDK limitation for non-English voice agents. Architected scalable telephony infrastructure using Asterisk as middleware to route calls to various endpoints including WhatsApp.
Built a full-stack AI backend orchestrating 8+ specialized agents for analysis, chat, forecasting, and image generation, serving 80+ onboarded cafes. Designed recommendation architecture and customer intelligence (RFM, churn). Production infrastructure with PostgreSQL, Redis, S3, and full async observability via Logfire.
Led end-to-end planning and execution of a multi-day flagship fest with 600+ participants and an 80-person team. Closed ₹21 lakh in sponsorship in under a month, built and deployed the event website, negotiated the first private VLAN on campus for King of the Hill infrastructure, coordinated speaker onboarding, and managed cross-functional volunteer teams. Inaugurated Recon on 19 May, two ECE lab exams on the 17th, Y Combinator Startup School in Bangalore on the 18th (met founders, product and validation discussions, afterparty with VCs), then flew back to open the fest.
Leading 100+ student developers in open-source practices. Conducting workshops on WebRTC, API Design, and Backend Development. Overseeing 6 open-source projects with technical roadmap and code review processes.
Full-stack CRM with AI-powered voice agents for inbound/outbound calls, Google Calendar integration, and real-time lead management via WebSockets.
Multimodal AI backend for a humanoid robot, vision transformers, hybrid memory with pgvector, real-time speech via WebRTC, and facial recognition.
Enterprise call log processing with self-hosted Whisper v3 transcription, PII sanitization, and multi-agent analysis for generating reports and insights.
Graph-based RAG system using Neo4j for knowledge retrieval with agentic reasoning, deep research via DuckDuckGo, and visualization of reasoning paths.
AI-powered postal mail sorting using OpenAI Vision for address extraction, India Post's Pincode API for smart routing, and Cloudflare R2 for storage.
Backend for the Ministry of Corporate Affairs, legislative sentiment analysis with PDF upload, AI summarization, and batch CSV processing using GPT-4.1.
Email or LinkedIn for roles, collabs, and interesting problems.