LOADING EXPERIENCE
SYSTEMS · MUSIC · EDUCATION · YOGA
PORTFOLIO
SAMEER KUMAR DASH
Voice agents · Semantic search · Media infrastructure · Automation
I’ve shipped 30 production systems across AI, voice, search, and media infrastructure. If you’re building in any of these — I’ve already solved problems like yours.
[email protected]Identify the friction. Remove it. Build something the next person can maintain without you. That method has guided every system I have shipped — from a recruiter's desk in Hyderabad to an AI lab in Coimbatore. The technology changed at every stage. The discipline did not.
Production AI: voice agents, semantic retrieval, and multi-modal search workflows.
Engineering leadership: end-to-end delivery, architecture, and operational automation at scale.
Infrastructure mindset: reliability, cost constraints, and clear boundaries that survive handoff.
Production Systems Shipped
Years Building
Industries Touched
Technology Domains Deep
PROFILE SNAPSHOT
15 Years of Shipping What Needed to Exist
TECHNICAL STRENGTHS
PRODUCT & DELIVERY STRENGTHS
SCROLL TO EXPLORE BY TECHNOLOGY
8 PRODUCTION SYSTEMS
From semantic search to voice agents. Production AI, not demos.
AI that measures empathy. Built for corporate coaching at scale.
EXPERIENCE·2024-2025 · Confidential AI Client (NDA)
AI-driven roleplay conversations that score empathy, detect triggers, and deliver coaching insights to team leads.
First measurable empathy tracking system for the client
Zero vendor costs. Voice AI that speaks the world's languages.
EXPERIENCE·Current · Open Source, In Development
Replaced all paid STT/LLM/TTS dependencies with local models on a single VM — enabling voice AI in languages no vendor supports.
Zero per-call API costs, supports low-resource languages
Made 2M photos searchable by meaning, not filename.
EXPERIENCE·Feb 2024-May 2024 · Isha Foundation AI Team, Coimbatore
Type what you're looking for in plain English. Find it across 2 million photographs instantly.
75% faster image selection across 2M photos
Ask a question. Hear the answer in a cloned voice.
EXPERIENCE·Jan 2024-Feb 2024 · Isha Foundation AI Team, Coimbatore
Built a voice-first RAG assistant — retrieval from teachings, reasoning with GPT, response spoken in a cloned voice.
First voice-first conversational interface for the archive
Search by meaning, not just keywords. Jump to the concept.
EXPERIENCE·Dec 2023-Jan 2024 · Isha Foundation AI Team, Coimbatore
Upgraded transcript search from keywords to semantic meaning — find concepts, not just words, across thousands of talks.
Intent-level retrieval across thousands of hours
Saved INR 1L+/month on dubbing. Trained the team to use it.
EXPERIENCE·Oct 2023-Dec 2023 · Isha Foundation AI Team, Coimbatore
Evaluated, deployed, and operationalized AI voice generation for publication-ready dubbing — saving over INR 100K monthly.
INR 1L+/month saved on dubbing costs
One prompt. Three layers. Complete thumbnail.
EXPERIENCE·2024 · Colaba AI (POC for Gulliver Creator Thumbnail flow)
AI-generated 3-layer thumbnail assembly — background, text, character — composed automatically from a single prompt.
Thumbnail creation reduced from hours to seconds
Testing the frontier before betting production on it.
EXPERIENCE·Ongoing · Personal / Exploratory
Rapid experimentation on emerging AI capabilities — CAPTCHA bypass with Vision API, Whisper transcription pipelines, and more.
OPEN SOURCE
OPEN-SOURCE · VOICE INFRASTRUCTURE
LiveKit is an open-source, WebRTC-based framework for building production-grade voice and video AI agents. It has become the infrastructure layer of choice for the AI industry — enabling developers to deploy real-time conversational agents where a human speaks, the system listens, reasons, and responds, all within sub-500 millisecond latency windows.
A LiveKit voice agent is, technically, a participant in a real-time media room. It processes audio through a three-stage pipeline: Speech-to-Text (STT) converts the user's voice into words; a Large Language Model (LLM) interprets meaning and generates a response; Text-to-Speech (TTS) renders that response as spoken audio. The entire exchange happens in the time it takes a human to pause and think.
SKILL PROGRESSION
2011
Mainframe Foundations
2010-2012
Workflow Automation
2012-2015
Full-Stack System Design
2017-2020
Enterprise Media Architecture
2020-2022
Automation at Operational Scale
2023-2025
AI Platform Strategy
SOUNDS OF ISHA
This section is designed as seven sonic chapters. Every chapter holds context first, then the song, so each video feels like part of one cohesive body of work.
A movement from restlessness to centeredness, written like a reminder that inner balance is not an idea but a lived discipline.
SONIC CHAPTER
A devotional invocation carried by rhythm and surrender, composed to hold stillness and intensity in the same breath.
SONIC CHAPTER
An ode to those who leave their soil to survive, and still carry the scent of home inside every unfinished sentence.
SONIC CHAPTER
A night-song of fire and celebration, made to echo collective devotion with urgency, drums, and joy.
SONIC CHAPTER
A prayer in melodic form, shaped around humility, gratitude, and the tenderness of offering oneself fully.
SONIC CHAPTER
A powerful chant of reverence and force, written to feel like a procession where each line rises like a call.
SONIC CHAPTER
A reflective spoken meditation on origin, awareness, and the intelligence that silently holds all creation together.
SONIC CHAPTER
AI LABS: THE LYRICIST'S EXPERIMENT
These are AI-assisted songs where the lyrics remain deeply personal while the arrangement, voice textures, and tonal experiments are built with generative tools. This is not replacement. This is a new instrument.
"A contemplative journey where devotional lyrics meet AI-crafted sound design."
Written from an inner quest, this track explores surrender, inquiry, and movement through layered electronic textures.
"A gratitude hymn reinterpreted through modern production and AI voice experimentation."
This composition carries the spirit of devotion while testing how machine-generated tonal spaces can hold emotional depth.
"A fierce, immersive chantscape inspired by Bhairavi energy and ritual rhythm."
Structured like a sonic invocation, this song blends lyrical intensity with synthetic atmospheres and percussive drive.
"A soft-burning melody on longing, slowness, and emotional afterglow."
This piece experiments with minimal AI orchestration to keep the lyric intimate while the arrangement breathes in quiet layers.
FIELD WORK
Teaching where students are, not where systems expect them to be.
I have taught students to build, communicate, and complete real work. From Isha Vidya classrooms to village classes in Uttarakhand, the focus is the same: make learning practical, make it accessible, and make it complete.
ISHA VIDYA
I did a project with the Isha Vidya students where I was teaching them entrepreneurship through making websites.
I also gave them training on how they can approach different customers and offer them a website, and I taught them how to deploy a website so they can complete the full customer process.
UTTARAKHAND
Apart from that, I have taught village kids in Uttarakhand, in Rudraprayag district.
In the village gates, I would gather them and conduct special classes so they would be able to learn English.
TEACHING METHOD
I am deeply passionate about teaching software. One technique I have used is poetry — translating programming concepts into rhythm and imagery so students can remember ideas with ease.
POEM PREVIEW
Devi, help me do my best. As finding a nameless variable is my quest. As I look beyond the value of a variable, the name is there, pointing to a memory block that the system can spare.
Devi, help me do my best. As finding a nameless variable is my quest. As I look beyond the value of a variable, the name is there, pointing to a memory block that the system can spare. As I wander around the declarations, sieving through the fat data-filled and starving valueless variables, all of them have a type, all of them have a name. As I peek in the three conditions of for, looping on in the code block, the variable initiates, changes, and finalizes; a name still represents the changing stock. In the options given by if and else if, and in the inevitability of else, I find a true and false, but it's just a Boolean clause. While this happens, do this; while that happens, do that. I play the block continuously, but I can't find the sound of a nameless variable in this syntactic cacophony. I stumbled on a for each, for each variable and then came a name. Some said "as a name," some told the name in; still, the name was there. My pessimism says I've lost this game. To get the desired action, I planned to make my own function. I wrote the function keyword and a name; the parameters in the bracket were also names. The parameter names showed up inside the function too. Only if they were born nameless and grew nameless, it would have been cool. The function then returned a value; I thought, this is it. But then I was visited by VBScript and JavaScript: the returned value has a name; the function name was the same. They said everything is either an object or is found inside an object. Then who am I to object? So I created an instance of an object, or may I say the object was set. And then I put a dot and tried to add a nameless variable into my plot. Alas, fate was cruel: objects only allow names after the period, any one among the dual - name of a function or variable. Then I found an array, collection of items that don't stray. These are my nameless variables, I declare. But each of them had a key or number for them, which the array had uniquely spared; that was their name. Made me feel like I'm lame, what a structural shame. But I wasn't done with arrays. I decided to look into each key, each number; maybe they were my nameless ones. But the sky broke and foretold: you will fail, as they are fields and not variables, as thought by some. In my helplessness I called my friend, the for. And she declared a variable so humble, so humble that it started at zero and stepped one number in each loop. It kept going, stopping at a number less than the array contained items. I put brackets and wrote the number beside the array name; they popped out values, but still no nameless variable to blame. In my quest I faced many errors, trepidation followed by shivers. To locate my foe, I planned ad hoc: a try and catch block. I wrote a print message in catch block to catch errors in exception object and mock. It did its job, showed errors round the clock. No such thing as a nameless variable. No, it exists not.
WHAT'S NEXT
I am presently working on creating an application that students can use through WhatsApp to further their learning and clear their doubts.
They will ask a question in voice or text, and it will send back a video that contains the answer, like a video presentation.
This is one of the things I wish to do for education: helping students learn easily by leveraging AI, especially for students who are not good in English, so they can further their learning and learn to speak better English.
PRODUCT FLOW
Student asks a doubt on WhatsApp in voice or text -> AI understands context -> student receives a video answer that explains the concept clearly.
"I am building this so students can ask in their own language, learn with confidence, and grow without language being a barrier."
FIELD WORK
Teaching where students are, not where systems expect them to be.
For eight years at Isha Ashram in Coimbatore, I followed a daily schedule of yoga and sadhana alongside contributing my skills to the foundation's work. That rhythm taught me something no sprint or deadline ever could: that meditation, like salt in a dish, doesn't need to take up space to be essential. It is present, invisible, and indispensable.
FOCUS
Yoga gave me the ability to walk into high-stakes situations without losing a single percent of my attention. Not calm in spite of the chaos — calm as a deliberate choice within it.
PERSPECTIVE
Eight years of practice opened something unexpected: the ability to genuinely understand people whose worldview is completely opposite to mine. Not just tolerate — understand. That made collaboration possible where conflict used to live.
PRESENCE
This is what I learned. Not a pose, not a technique — a shift in where I place my awareness. When that became real, it changed how I lead, how I listen, and how I work.
PLACE
Home for eight years. Yoga, sadhana, and the daily work of contributing technological skills to one of the world's largest foundations.
EXPLORE THE PROGRAM
PLACE
A sea of seekers, each on a different path, sharing the same river. Proof that people can grow together even when they start from completely opposite places.
PLACE
The Himalayas don't just inspire — they recalibrate. The vibe of a place is as important as its decor. I travel to these spaces not to escape, but to deepen.
PLACE
A quieter Himalayan village, less visited, more felt. Some of my clearest perspectives have arrived in these overlooked meditative spaces.
PLACE
Home for eight years. Yoga, sadhana, and the daily work of contributing technological skills to one of the world's largest foundations.
EXPLORE THE PROGRAMPLACE
A sea of seekers, each on a different path, sharing the same river. Proof that people can grow together even when they start from completely opposite places.
PLACE
The Himalayas don't just inspire — they recalibrate. The vibe of a place is as important as its decor. I travel to these spaces not to escape, but to deepen.
PLACE
A quieter Himalayan village, less visited, more felt. Some of my clearest perspectives have arrived in these overlooked meditative spaces.
READING
My favorite book. Osho's commentary on this ancient tantric text is the most practical guide I have found for understanding different forms and methods of meditation — and how they can be incorporated quietly into daily life. The two volumes together are a complete map.
Not every system I have built began as a job. Some began as frustration. Some as curiosity. Some as the quiet recognition that something useful ought to exist in the world, and no one had yet made it. This section holds those — the experiments outside the commission, the products built in the margins of time, the architectural decisions made not because someone asked for them but because the problem was real and the solution was within reach.
PRODUCTS IN PROGRESS
These are not side projects. They are products built from the conviction that the problem is real, the technology is ready, and the gap exists because no one has sat down long enough to close it. Each one was designed in the margins of time — evenings, weekends, the hours between one assignment and the next. Each one addresses a specific failure of the existing landscape. They are at different stages of completion. They will be deployed when the conditions are right.
ACTIVE · BETA LIVE
WHAT IT DOES
Stonks Radio turns stock tracking into a live audio experience. A user picks any NSE stock, presses play, and receives AI-generated commentary in simple language with updates every three minutes during market hours. The product is built for real use: no app download, browser-first playback, and support for English, Hindi, and Assamese.
The stream focuses on clarity, not noise — price movement, volume shifts, sector pulse, and key market context explained in plain speech. It is designed to fit into a commute, a workday, or a short break, so users can stay informed without being locked to charts all day.
THE GAP IT CLOSES
Retail traders are forced into high screen-time workflows: constant app checking, fragmented YouTube/news inputs, and jargon-heavy analysis that increases anxiety instead of improving decisions. Existing tools optimize for visual engagement. Stonks Radio closes the gap with low-friction, audio-first intelligence built for attention, accessibility, and disciplined decision support — without buy/sell calls.
Live in beta with browser-based access and WhatsApp sign-in.
stonksradio.liveIf you are an investor, co-founder, or early adopter who sees what these products are trying to do — the conversation is open.
[email protected]