Open to Forward Deployed and AI Engineer roles

I build AI agents that do real work.

Voice agents that answer the phone. Operations agents that work the inbox and ask before they act. And the open-source infrastructure that makes them safe to ship.

AI engineer · Founder, MagicTeams.ai · Hyderabad, India

See what I build GitHub ↗LinkedIn ↗

AIOS desktop showing an outcome-first morning brief

AIOS on iPhone asking the owner to approve an action

AIOSAgents do the work. You approve what matters.

<400 msto respond on live phone calls

33 branchesof a logistics company on one agent inbox

57 mailboxessynced into a single sales pipeline

5 MCP serverssubmitted to Meta's Muse directory

What I build

Three kinds of software, one idea underneath.

Agents should take real work off a team's plate, stay inside clear limits, and leave proof of everything they did.

01 Voice agents

Phone agents that answer, qualify, and book.

Speech-to-speech over real telephony, not a demo widget
Handles interruptions mid-sentence, like a person would
Responds in under 400 ms on live calls

Built: MagicTeams.ai voice platform, Magic Voice Teams

Live call · Home services line<400 ms

CallerHi, our AC stopped working this morning.
AgentSorry about that. I can get a technician to you tomorrow between 9 and 11. Does that work?
Caller · interruptsActually, is there anything today?
AgentThere is a 4 pm slot today. I have booked it and texted you a confirmation.

Illustrative call · books into the calendar · logs everything

02 Operations agents

Agents that work the inbox, CRM, and books.

Reads email, WhatsApp, CRM, and finance data in one place
Opens with money at risk, not another dashboard
Every outbound action waits for a person to approve it

Built: AIOS, AIOS for Home Services, client systems in logistics and manufacturing

AIOS for Home Services showing revenue at risk and recovered revenue — AIOS for Home Services
Missed calls, stale estimates, and dormant leads, ranked by money.

03 Agent infrastructure

The plumbing that makes agents safe to ship.

MCP tools with typed schemas and approval tokens
Memory and business context that carry across sessions
CI and coordination when many agents share one codebase

Built: Flare, agentprofile, Distillory, Primer, Agent Ready

agent.tracetyped · approved · replayable

10:41:02 tool     crm.find_customer { name: "Ravi Freight" }
10:41:03 tool     rates.price_route { from: "HYD", to: "PNQ" }
10:41:05 policy   email + money => needs approval
10:41:05 ask      Send ₹48,000 quote to Ravi Freight?
10:42:17 approved by owner, on mobile
10:42:18 tool     mail.send_quote { token: "apv_7f3c" }
10:42:19 receipt  3 tools · trace saved · undo for 10 min

Illustrative trace · the same shape every agent leaves behind

How it works

Products that work. Four nouns, everywhere.

Every system I ship uses the same model, so a team always knows what the agent is doing, what it is waiting on, and what it already did.

Each agent watches one part of the business. It can read what it needs and nothing more.

Illustrative example of a freight enquiry moving through AIOS.

New enquiry12 tonnes, Hyderabad to Pune, pickup FridayOps agent picked it up from the freight inbox

In production

Running in real businesses, not just demos.

Client names are left out where the work is private. Happy to walk through any of it on a call.

LogisticsIn production

A 33-branch logistics company

A unified inbox and opportunity engine. 57 mailboxes flow into one pipeline that goes from enquiry to quotation to truck to trip.

VoiceIn production

MagicTeams.ai voice platform

Speech-to-speech phone agents with telephony, cloned voices, and interruption handling, answering real business calls in under 400 ms.

ManufacturingFixed-fee pilot

A Nevada solar recycler

A production run log with mass balancing across staged, WIP, and offtake, certificate issuance, and badge and PIN sign-in for operators.

Go-to-marketCTO, part-time

AcquisitionX

From LinkedIn engagement to qualified leads: ICP research, drafted outreach with approval and dedupe, synced into Smartlead campaigns.

Flagship product

AIOS

A durable operating layer for a company. Desktop keeps local work private, the cloud keeps jobs moving, and mobile keeps the founder in control.

AIOS Mobile

Brief. Needs you. Ask. Pulse. The remote control for AIOS on iOS and Android, without squeezing the desktop onto a phone.

Download AIOS ↗Release notes ↗

Open source

The tools I build agents with, published.

All public on GitHub. Most run on Cloudflare and speak MCP.

CI · Cloudflare

Flare

CI and coordination for many AI agents on one repo, on your own Cloudflare.

Identity · MCP

agentprofile

One agent identity, skills, and memory, synced to every tool through MCP.

Memory · SQLite

Distillory

A local-first memory engine stored in one inspectable SQLite file.

Context

Primer

Gives a new agent session the business context it should have started with.

MCP · REST · CLI

Agent Ready

Turns products into safe MCP, REST, and CLI surfaces for agents.

Workspace

PlainSync

A shared Markdown workspace where people and agents write together.

Research

Muse use cases

351 real-world agent use cases, 257 with the exact prompt.

macOS

Notch Limits

Claude and Codex subscription limits, visible from the Mac notch.

More work

More things that made it out of my head.

Products, client work, and experiments. The range is the point: learn a new world quickly, find the real outcome, ship something people use.

AI platform · 2024

EveryAI

Discover and use text, image, audio, code, and research models in one place.

Voice AI · 2024

Magic Voice Teams

Build, test, and deploy voice agents in minutes rather than months.

Research product

Idea Atlas

1,847 startup opportunities, organized from evidence.

Agent prototype

Agentic Inbox

An email agent that drafts, shows its work, and waits for a person to send.

Client delivery

Camp Revolution

A clear, accessible home for a mission that deserves one.

Glow

Routine and skin analysis

Alarm Maker

Custom alarms and challenges

Swing Coach

Record, score, improve

TallTrack

Finds the few ideas in a sales call worth publishing

Callcraft

Grounded writing from real sales calls, from the terminal

Lazystudio

Record lazily, then turn it into a polished video

I'm Phanindra. I like making the whole thing.

The product, the interface, the code, the operating model, and the awkward details between them.

I started in data engineering at Accenture, building ML pipelines on Azure and Google Cloud. Then I founded a content agency, an AI practice, and MagicTeams.ai, shipping voice agents and operations agents for businesses in logistics, manufacturing, real estate, and services. AIOS came from the bottleneck I knew best: trying to run everything while still finding time to build.

Hiring

For hiring teams.

Open to Forward Deployed Engineer and AI Engineer roles. I take an agent from a customer's messy workflow to a system their team trusts in production.

Production voice AI
Speech-to-speech phone agents with telephony, cloned voices, and barge-in handling at sub-400 ms live-call latency.
Governed agent runtime
Typed tools with schema validation, human approval tokens, read/write separation, action ledgers, and replayable traces.
Deployed in real businesses
A unified inbox and opportunity engine live at a 33-branch logistics company, plus manufacturing and services pilots.
MCP and agent infrastructure
Five MCP servers submitted to Meta's Muse directory, and open-source agent tooling: Flare, agentprofile, Distillory, Primer.

Roles: Forward Deployed Engineer, AI Engineer, Applied AI
Based in: Hyderabad, India
Open to: Remote, or relocating to San Francisco or New York with employer-supported U.S. work authorization
Stack: TypeScript, Python, React, Cloudflare Workers, D1, Durable Objects, MCP, LLM orchestration, RAG, evals

MagicTeams.aiFounder and AI EngineerNov 2024 to now
AcquisitionXCTO and Partner, part-timeSep 2025 to now
Every AIFounder, AI and Automation EngineerApr 2023 to now
Web3 WibeFounderAug 2022 to Jan 2024
AccentureData Engineering Analyst, ML EngineerJul 2021 to Nov 2022
NIT SilcharB.Tech, Electrical Engineering2017 to 2021

Email me ↗LinkedIn ↗Profile for AI agents ↗

Let's build
something real.