ā±ļø 6 min read
Architecting Ultra-Low Latency Voice AI Agents with Pipecat & SileroVAD
How we reduced end-to-end voice turnaround time from 1.8s down to sub-500ms for natural, human-like telephone conversations.
š¤
Hello
Iām Milind Wadhwa, a Senior Software Engineer based in Gurgaon, India.
I specialize in building production-grade Agentic AI platforms and scalable distributed backend systems. I architect multi-channel intelligent agents across Voice, Chat, SMS, and WhatsApp ā backed by RAG pipelines, LLM orchestration, and MCP ā while also having deep roots in traditional backend engineering: REST APIs, async queues, and microservices in Python, Golang, and Java.
With over 7 years of engineering experience across Bidgely, Plivo, and Optum, I bridge the gap between robust backend foundations and cutting-edge AI ā building systems that are both production-ready and genuinely intelligent.
7+ Yrs
Engineering Experience
85%
Autonomous Resolution Rate
~70%
System Efficiency Boost
What I Do
Combining deep GenAI research, speech models, and enterprise-grade distributed backend architecture.
Agentic AI Platform Engineering
Architecting production-grade multi-channel AI agent platforms spanning Voice, Chat, SMS, and WhatsApp. Deep expertise in LLM orchestration with LangChain and Agno, multi-modal RAG pipelines, MCP tool integration, and model optimization via LoRA / PEFT fine-tuning on open-weights STT and language models.
LangChain
Agno
MCP
RAG
Vector DBs
Pipecat
Deepgram
ElevenLabs
LoRA Fine-tuning
SileroVAD
Distributed Backend & Microservices
Proven track record in architecting high-throughput, low-latency microservices using FastAPI, Flask, Golang, and Java (Spring Boot). Transformed synchronous SaaS platforms into asynchronous architectures powered by Apache Kafka and Amazon SQS, accelerating service response efficiency by ~70%.
Python
FastAPI
Golang
Java
Spring Boot
Apache Kafka
Amazon SQS
PostgreSQL
Redis
Clickhouse
Cloud Infrastructure & DevOps
Extensive experience deploying cloud-native architectures on AWS (ECS, EC2, CloudWatch, SQS). Automated containerized microservice deployments with Docker, Kubernetes, and OpenShift, establishing resilient CI/CD pipelines via Jenkins to ensure 99.9% platform availability.
AWS ECS / EC2
Docker
Kubernetes
OpenShift
Jenkins CI/CD
CloudWatch
Linux Systems
Observability & Team Leadership
Established full-stack monitoring, distributed tracing, and real-time anomaly detection with Splunk, Elastic APM, Dynatrace, and the ELK Stack. Passionate mentor who leads code quality initiatives, conducts thorough PR reviews, and guides junior engineers toward software excellence.
Splunk
Elastic APM
Dynatrace
ELK Stack
Rest-Assured
Cucumber BDD
Tech Mentorship
Work Experience
7+ years of driving innovation in Generative AI, cloud platforms, and distributed systems.
š±
Software Development Engineer - 3 @ Bidgely ā
06/2026 - Present
š Gurgaon, Haryana, India (Remote)
Conversational Energy Assistant Chat & Voice Agents
- Architecting advanced conversational chat and voice agents to analyze home utility bills, provide detailed energy insights, and suggest optimal billing rate plans.
- Developing LLM-driven agentic pipelines to deliver actionable recommendations for lowering electricity consumption and optimizing overall home energy usage.
- Integrating real-time voice and messaging frameworks for low-latency, natural dialogue systems in consumer utility engagement.
š¬
Software Development Engineer - 2 @ Plivo
Communications ā
01/2024 - 05/2026
š Bangalore, Karnataka (On-site)
Engineering Voice AI Agents & Multimodal Workflows
- Engineered low-latency Voice AI Agents utilizing Pipecat, SileroVAD, Deepgram, and ElevenLabs for natural voice interactions.
- Spearheaded model optimization using LoRA fine-tuning for STT models to significantly reduce transcription latency and improve accuracy.
- Led technical integration of multi-modal LLMs and Vector Databases, optimizing RAG pipelines for superior retrieval across WhatsApp, Webchat, and SMS channels.
- Architected and deployed a production-ready Agentic AI ecosystem leveraging LangChain and Agno to automate support workflows with 75% resolution accuracy.
- Mentored junior developers and conducted thorough PR reviews to maintain rigorous code quality and architectural standards.
š¬
Software Development Engineer - 1 @ Plivo
Communications ā
12/2021 - 12/2023
š Bangalore, Karnataka
SaaS Platform Refactoring & Cloud Scalability
- Revitalized SaaS platform performance by migrating synchronous services to asynchronous FastAPI and implementing task parallelism via Amazon SQS, boosting efficiency by ~70%.
- Developed core platform APIs utilizing Flask and Golang for high-volume sales engagement applications.
- Automated ECS cloud deployments and configured EC2 auto-scaling via CloudWatch for 99.9% system availability.
š„
Associate Software Engineer - 2 @ Optum
(UnitedHealth Group) ā
07/2019 - 12/2021
š Hyderabad, Telangana
Microservices & DevOps Automation
- Built RESTful microservices using Spring Boot and Apache Kafka for real-time, event-driven healthcare data streaming.
- Automated CI/CD pipelines and containerized deployments on OpenShift using Docker and Jenkins.
- Reduced test execution duration by 40% via multi-threaded integration testing with Rest-Assured and Cucumber BDD.
- Implemented full-stack monitoring using Splunk, Elastic APM, and Dynatrace to improve system observability.
š„
Software Engineering Intern @ Optum
(UnitedHealth Group) ā
05/2018 - 07/2018
š Gurgaon, Haryana
Observability & Anomaly Detection
- Deployed an ELK Stack solution for real-time anomaly detection, successfully monitoring over 100K email transactions for irregularities.
Technical Highlights
Key architectural solutions and production systems engineered throughout my journey.
Omnichannel Agentic AI Platform
Architected a full production agentic platform across Voice, Chat, SMS, and WhatsApp ā with RAG pipelines, MCP tool integration, and LLM orchestration via LangChain & Agno. Achieved an 85% autonomous resolution rate for customer support workflows.
STT Model LoRA Fine-Tuning
Applied parameter-efficient fine-tuning (PEFT) using LoRA on open-weights Speech-to-Text models, significantly reducing transcription latency and improving domain-specific accuracy with 75% less VRAM vs full fine-tuning.
Async SaaS Platform Refactor
Boosted platform efficiency by ~70% by migrating synchronous services to asynchronous FastAPI and offloading background work to Amazon SQS queues, with auto-scaled ECS deployments for 99.9% availability.
GenAI Energy Assistant
Building conversational chat and voice agents that analyze home utility bills, surface energy insights, and recommend optimal billing rate plans ā using LLM-driven agentic pipelines for real-time consumer engagement.
Get In Touch
Have an opportunity, a project, or want to connect about AI?
I'd love to hear from you! Feel free to reach out directly or send a message below.