Technical Lead, AI Voice Engineer

Open · UTC+05:30

HemendraTripathi

ARCHITECTURE BILLING TEAMS REVENUE

I build real-time voice AI that works on actual phone calls. I lead engineering for Callin.io, a voice platform serving 500+ businesses across healthcare and real estate, and I own everything from the call path to the infrastructure bill. Callin.io is pre-incubated by I3P, the incubator of Politecnico di Torino.

Udaipur, IN "भारत" · Remote from India, working with APAC time zones · Callin.io Tech Lead

businesses served
500+
lower LLM inference cost
~20%
fewer cancellations
~40%
(01)

Voice AI Case Study

Featured voice AI case study

Callin.io: building a voice AI platform that holds up on real calls

How I helped take Callin.io from first commit to a platform serving 500+ businesses across healthcare and real estate, with lower cost per call and fewer cancellations.

The problem
  • Every turn of a voice call is a race. If the agent takes too long to answer, the caller hangs up.
  • Using one large model for everything burned margin on simple turns like greetings and confirmations.
  • Customers were leaving, and the causes were spread across latency, voice quality, language support and how hard agents were to manage.
Architecture
Production voice pathCallin.io · live turn
  1. 01
    Caller
    Phone or web
  2. 02
    Telephony
    SIP · WebRTC
  3. 03
    Orchestrator
    Routing · cache · latency budget
  4. 04
    Models
    STT · LLM · TTS
  5. 05
    Post-call analytics
    Insights back to the customer

We own the path end to end. Cost and latency are visible on every turn.

What I built

Route by complexity

Classify each turn and send it to the smallest model that can handle it, keeping large models for hard reasoning. This is what cut inference cost by roughly 20%.

Treat latency as a budget

Each part of the call path gets a time allowance, and anything that blows it gets fixed first.

Choose providers on evidence

I compared speech and voice providers side by side and recommended a switch, which the CEO approved.

Learn from every call

Post-call analytics (goal analysis, sentiment, key moments, recommendations) show customers what to fix in their agents.

ReactNode.jsSupabaseStripeRedis
(02)

More AI Product Work

Callin.io is the deep dive above. These are related suite products and earlier shipped work. Shorter notes, same bar for outcomes.

01

CondoMail

Production
Product architecture: multi-provider email sync

AI agents that sort, draft, and auto-reply across providers. Live with early adopters running real inbox workflows on it.

React / Node.js / Supabase / Stripe / Firebase
02

Realead

Beta
Full-stack · mobile · AI calling flows

Connects lead sources, builds a business profile, and places AI qualification + follow-up calls for property leads. Final beta ahead of release, and the conversational model behind the agent demo.

React Native / NestJS / Supabase / StripeSee the demo →
(03)

Experience

FEB 2026 to PRESENT

Technical Lead, AI Voice PlatformAppspundit Infotech · Callin.io

  • Own the call path, multi-LLM routing, self-hosted Qwen for post-call analytics, and the AI-agent ticket pipeline.
  • Own architecture decisions and infrastructure spend for a platform serving 500+ businesses across healthcare and real estate.
  • Cut LLM inference cost roughly 20% by routing each request by complexity.
OCT 2024 to JAN 2026

Software Engineer, AI Voice PlatformAppspundit Infotech · Callin.io

  • Built Callin.io from the ground up: voice agents, carrier integrations, multi-tenant architecture and Stripe usage-based billing.
  • Extended the shared multi-LLM architecture into CondoMail and Realead.
2021 to 2024

Earlier: engineering and teachingAimers Institute

  • Built 10+ web applications across MERN, Django, Flask and Laravel.
  • Taught 150+ students Python, Django, MERN and Flutter through project-based learning.
BCA, Mohanlal Sukhadia University (2022)English: fluent · Hindi: native
(04)

The Demo Is the Résumé

I build AI voice agents that make real phone calls for a living. This one runs on the same conversational patterns as production voice agents, except its lead-qualification target is you.

Prefer the short path? Read the case study. Prefer proof? Answer the call. The lead file builds as you ask.

  • → Same patterns as production qualification agents
  • → Finish the call. Get the summary. Email if it lands.
(05)

How I Think About Voice AI

01

Latency is the product

In voice AI, silence is a bug. Every architectural choice (caching, routing, carrier failover) exists to keep the human from hanging up.

02

Pay for intelligence only when you need it

A confirmation doesn't deserve a frontier model. Route by complexity. Your CFO will notice. So will your p95.

03

Billing that doesn't create tickets

If customers argue about invoices, the product is unfinished. Usage ledgers should be boringly correct.

04

Own the stack's P&L

Architecture without vendor spend ownership is theater. I ship the system, and I know what the infra bill was last Tuesday.

(06)

AI & Engineering Capabilities

Voice AI Systems

  • Real-time telephony: SIP/VoIP and WebRTC
  • STT/TTS selection, streaming and latency budgets
  • Multi-LLM routing by complexity and cost
  • Self-hosted Qwen for post-call analytics

Product & Revenue

  • Usage-based billing architecture
  • Stripe subscriptions & invoicing
  • Pricing design & infra cost optimization
  • Client acquisition & retention

Engineering

  • React, Next.js, React Native
  • Node.js, NestJS, Python
  • PostgreSQL (Supabase), Redis
  • Event-driven systems and REST APIs

Cloud & Leadership

  • Docker, AWS, CI/CD pipelines
  • System design and architecture decisions
  • Technical leadership and mentorship
  • Remote, cross-timezone collaboration
(07)

Off the Record

Hemendra Tripathi, AI Voice Engineer and Technical LeadHemendra Tripathi in a playful Batman portrait

Based in Udaipur. Remote from India, working with APAC time zones. I got here by teaching 150+ students to code and rebuilding a voice AI platform until 500+ businessesrelied on it. I like systems that are fast, boring and profitable, and I’m drawn to how people think and decide, which shapes how I design agents that talk to them.

(08)

FAQ

Who is Hemendra?

Hemendra is Hemendra Tripathi, an AI Voice Engineer and Technical Lead based in Udaipur, India. He leads engineering on Callin.io, a voice AI platform serving 500+ businesses, and works fully remotely.

Who is Hemendra Tripathi?

Hemendra Tripathi is an AI Voice Engineer and Technical Lead who builds real-time voice agents. At Appspundit Infotech he joined Callin.io as a Software Engineer in October 2024 and became Technical Lead in February 2026. He holds a BCA from Mohanlal Sukhadia University.

What is an AI Voice Engineer?

An AI Voice Engineer designs and ships real-time voice agents: telephony, speech-to-text and text-to-speech, language-model routing, latency and cost. Hemendra Tripathi does this in production at Callin.io.

Which Hemendra Tripathi is the Callin.io technical lead?

The one who leads engineering at Callin.io is an AI Voice Engineer based in Udaipur, India, and the author of this site. He is not the journalist or the academics who share the name. His LinkedIn and GitHub are linked from this page.

How do I hire an AI Voice Engineer like Hemendra?

Book a 20-minute call from the hire page. Hemendra is open to fully remote Applied AI, voice-agent and forward-deployed engineering roles, and replies within 24 hours.

What is complexity-aware multi-LLM orchestration?

Each turn of a call is classified and sent to the smallest model that can handle it, with large models reserved for hard reasoning. At Callin.io this cut LLM inference cost by roughly 20%.

What does Hemendra work on?

Real-time voice systems (SIP/VoIP and WebRTC, speech-to-text and text-to-speech, latency budgets), multi-LLM routing and post-call analytics, plus the React, Node.js, Supabase and Stripe platform around them.

What results did the Callin.io voice AI platform achieve?

Callin.io serves 500+ businesses across healthcare and real estate. Complexity-aware routing cut LLM inference cost by roughly 20%, and cancellations fell by roughly 40%. The case study explains how.

What roles is Hemendra open to?

Fully remote Applied AI, voice-agent and forward-deployed engineering roles, and technical-lead roles. Remote from India, working with APAC time zones.

(09) Contact

HIRE ME.

Open to fully remote AI Voice Engineer, applied AI and technical-lead roles. Pick a 20-minute slot (timezone is converted for you). Replies within 24 hours.

Remote from India, working with APAC time zones