Mahesh Sonker
Built to be Everyone

Twenty years of technology. Now exploring humanity.

Note — A personal portfolio of self-initiated research, learning, innovation, technical exploration, and my personal aspiration to serve humanity through technology

Two decades of building technology.
Now dedicated to understanding humanity.

Exploring AI that understands your emotions, memories, behaviors, empathy, and decisions—creating your personal digital avatar that thinks with you, grows with you, and stays connected with your loved ones.

Your Voice. Your Memories. Your Digital Self.

Mahesh Sonker — Technology Leader
20+ years in technology
Problem Solver Tech Veteran Scaling Specialist AI Explorer Innovation Lead
— Profile

Mahesh Sonker

Technology Leader · AI Explorer

20+ years Gurugram, IN Thinking
About

More than a career.
A way of thinking.

Technology has been more than a career for me — it has been over two decades of solving complex problems, building scalable platforms, leading high-performing engineering teams, and continuously learning through innovation.

As a technology leader with 20+ years of experience, I've been fortunate to architect and scale large-scale e-commerce, loyalty, and enterprise platforms — transforming ideas into reliable, high-impact products. I enjoy simplifying complexity, driving engineering excellence, and building technology that delivers measurable business value.

While my foundation lies in scalable software engineering, my curiosity has led me toward the next frontier of AI. Today, I am deeply exploring Emotional AI, Metacognition, AI Companions, Digital Clones, Voice Cloning, and Human-Centric Intelligence — building systems that go beyond data processing to understand human behavior, emotions, memories, and decision-making.

I am particularly fascinated by the convergence of voice intelligence, behavioral modeling, memory architectures, and emotionally adaptive AI — with the vision of creating digital systems that feel natural, empathetic, and deeply personal.

I believe the future of AI is not defined by larger models alone — it will be shaped by systems that understand people, adapt with experience, reason about their own decisions, and augment human potential.

— Belief
Twenty years building products. The next chapter: technology that quietly helps people through life's hardest days.

Free, accessible, human — by design

Focus areas

What I'm Exploring...

Five focus areas converging on a single idea: AI built around how humans actually think, feel, and decide.

01 / FOCUS Featured

Emotional AI

Systems that read tone, context, and the unspoken weight behind a sentence — AI that feels less like a query engine, more like a thoughtful collaborator who notices when you're tired, frustrated, or finally onto something.

Empathy models Tone & sentiment Adaptive response Emotional memory Affective computing
02

Metacognition

AI that reasons about its own reasoning — knowing what it doesn't know, when to ask, when to slow down, when to revise.

Self-awareness Uncertainty Reflection
03

Digital You

A faithful digital counterpart — voice cloned, memory captured, behavior modeled. One that can speak, decide, learn in your stead. Long after you've stepped away.

Voice cloning Personal AI Behavioral patterns Memory snapshots
04

Human-Centric AI

AI built around how humans think, feel, and struggle. Free, accessible technology that meets people where they are — especially when life is hardest.

Free & accessible Empathy-first design Augmentation Human-AI symbiosis
05

Voice Intelligence

The most natural human interface — finally adopted by machines. Cloning, synthesis, and conversation that knows when to listen.

Voice cloning Real-time STT Conversational AI
The stack

Technologies behind the curiosity.

Engineering depth meets AI exploration — decades of platform work underpinning every new prototype.

01

Foundation Models

Reasoning · Language
Claude GPT-4 / 5 Gemini Llama 3 Mistral Fine-tuning
02

Voice & Speech

Cloning · Synthesis · ASR
ElevenLabs Whisper Coqui TTS PlayHT Custom voice models Real-time STT
03

Memory & Context

Vectors · Retrieval · RAG
Pinecone Weaviate pgvector LangChain LlamaIndex Knowledge graphs
04

Behavioral Layer

Modeling · Adaptation · Personalization
Reinforcement learning Sentiment analysis Multi-modal models Custom embeddings User-state graphs
05

Engineering Spine

The 20-year foundation
Node.js Python Java React / Next.js AWS · GCP Microservices Kubernetes PostgreSQL · Redis Event-driven systems
"

The future of AI is not defined by larger models alone — it will be shaped by systems that understand people, adapt with experience, and reason about their own decisions.

Experience

Twenty years of shipping,
before any of this was called AI.

Engineering excellence isn't a buzzword on a slide — it's a habit built over two decades of architecting, scaling, leading, and delivering.

0+ Years architecting and scaling enterprise platforms SINCE THE EARLY 2000s
0M+ Users served across loyalty and e-commerce systems CUMULATIVE REACH
0+ Engineers led across high-performing teams TECH LEADERSHIP
Curiosity for what comes next in Human-Centric AI AND STILL COUNTING
Career

Twenty-two years of building.

From Java games at Walt Disney to leading engineering at Asort — seven companies, five mobile platforms, two startups, and one constant: shipping at scale.

2019 — Present

Asort.com

Current Role

Director of Engineering

Gurugram, Haryana, India

  • Designed scalable multi-vendor marketplace architecture leveraging microservices, AWS, and Docker containerization
  • Maintained 99.9%+ platform uptime through robust monitoring, alerting, and infrastructure engineering
  • Integrated AI/ML models for advanced analytics and business intelligence — enabling data-driven strategies across departments
  • Championed full-stack MERN adoption (MongoDB, Express, React, Node) and cultivated a high-performance, data-driven engineering culture
  • Leveraged Python automation to enhance development efficiency and data interpretation
AWS Microservices Docker MERN Python AI/ML DevOps
2018 — 2019

IFFCO iMandi

Sr. Engineering Manager

Gurgaon, India

  • Spearheaded AI-powered product discovery using neural networks for image recognition and NLP for voice search
  • Drove a 20% increase in user engagement and conversion rates
  • Reduced critical production bugs by 80% through rigorous QA and data-protection processes
  • Established streamlined stakeholder feedback loops translating business requirements into technical roadmaps
Neural Networks NLP Computer Vision AI/ML
2016 — 2018

Mesmerise Solutions – A Salon Booking App

Co-Founder

Co-Founder & CTO

  • Architected and led development of the "Mesmerise" social mobile app from concept to launch on Android and iOS
  • Built native experiences using Java/Kotlin on Android and Objective-C/Swift on iOS, with Firebase powering real-time features and analytics
  • Integrated novel technologies for a unique user experience — yielding competitive advantage that led to acquisition
Android iOS Kotlin Swift Firebase Founding Team
2014 — 2016

Mobibeans Solutions

Co-Founder

Co-Founder

  • Designed core architecture for "YelaTalk" — a real-time chat application across Android and iOS
  • Implemented custom protocols over WebSockets for efficient real-time messaging at scale
WebSockets Android iOS Real-time Systems
2006 — 2014

RockeTalk

8 years · 5 roles

Senior SWE → Group Lead → Tech Lead → Tech Manager → Engineering Manager

New Delhi Area, India

  • Eight-year progression through five engineering leadership roles — from individual contributor to managing teams
  • Architected RockeTalk mobile applications across Android, iOS, Windows Phone, and Symbian during the mobile revolution
  • Led system design, supervised engineering teams, and acted as the key technical resource aligned with business strategy
  • Designed for emerging smartphone platforms — porting and re-architecting for each new generation
Android iOS Windows Phone Symbian System Design Team Leadership
2005 — 2006

The Walt Disney Company

Software Engineer

Mumbai, Maharashtra, India

  • Ported enterprise mobile games across a variety of handsets distributed globally
  • Worked on Disney's global mobile gaming portfolio during the J2ME era
J2ME Mobile Games Cross-platform Porting
2004 — 2005

Handygo Technologies

First Role

Game Developer

New Delhi

  • Started career as a J2ME game developer at the dawn of mobile gaming
  • Built games for Nokia Series 30 and Series 40 — the dominant mobile platforms of the era
  • Recognized as Star Debutant Programmer for outstanding contribution as a new entrant
J2ME Nokia S30/S40 Mobile Games
Education

Always learning.

From foundational computer science to recent AI/ML and data engineering — credentials earned across three of India's notable institutions.

2021 — 2023

MTech, Data Science & Engineering

Birla Institute of Technology and Science, Pilani

An all-India Institute of Eminence since 1964. Multi-campus excellence in engineering, sciences, and management — where innovation and leadership go hand-in-hand.

Recent · Post-Graduate
2018

AI & Machine Learning

International Institute of Information Technology, Hyderabad

A premier autonomous research university established 1998. Leading India's research in computer science, AI/ML, language tech, and emerging digital sciences.

Certification · ML & Deep Learning
2000 — 2004

BTech, Computer Science

Dr. A.P.J. Abdul Kalam Technical University

One of India's largest technical universities, named after the Missile Man Dr. APJ Abdul Kalam. Affiliating hundreds of engineering colleges across Uttar Pradesh.

Foundation · Undergraduate
Currently

Active explorations.

Where my time and thinking are going right now — building, researching, and questioning.

Building

A voice-first Digital You prototype.

Cloned voice, behavioral patterns, and a small memory architecture — learning to speak, decide, and remember the way one person does. Currently teaching it when not to answer.

Researching

Metacognition in compact models.

Can a 7B-parameter model meaningfully reason about its own confidence — say "I'm not sure" instead of hallucinating? Early experiments suggest yes, with the right scaffolding.

Thinking about

How AI can quietly help people who struggle the most.

Mental health, isolation, accessibility, language barriers, navigating bureaucracy — the most valuable AI won't be the loudest. It'll be the kind that someone needed and could actually reach.

Explorations

Small experiments. Big questions. Meaningful discoveries.

A growing collection of prototypes, sketches, and field notes — each pursuing one corner of the human-AI question.

EXPLORATION · 01 Research

The Emotional · Wheel

A proven tool to identify how you feel — the first step in regulating your emotions. Used by psychologists, therapists, and anyone building emotional intelligence..

🔍 Identify Emotions · 🌱 Cultivate Awareness · ⚡ Boost Emotional Intelligence · 🧠 Expand Emotional
EXPLORATION · 02 Research

Voice Cloning · Hinglish

A guide to cloning voices in Hindi-English — preserving cadence, breath, emotion, and the ethical frame around it. Preserving your authentic cadence — the pauses, the breath, the Hinglish rhythm — is what makes a clone feel like you, not just a generic voice. A practical, open-source-first guide.

🎯 True Tone · 🥁 Natural Rhythm · ⏸️ Authentic Pauses · 🎼 Expressive Prosody · 🗣️ हिन्दी–اردو Blend
EXPLORATION · 03 Research

Beyond the Algorithm

What happens when intelligence becomes aware of itself?

🧠 Metacognition · 🔮 Self-Awareness · 🔁 Recursive Improvement · 💓 Emotional Integration · ⚡ Consciousness
EXPLORATION · 04 Concept

Digital You

One personalised digital companion for every citizen of India — multilingual, profession-adaptive, age-inclusive, built on open networks and India Stack..

🗣️ 22+ Languages · 👨‍🌾 Every Profession · 👶👴 All Ages · 🇮🇳 India Stack
EXPLORATION · 05 Concept

Cognitive Companion

A Digital Companion does not just respond — it feels the shape of your mood, reflects on its own reasoning, and grows with you across time. It is memory, empathy, and intelligence — fused.

🧠 5-LLM Fusion · 💓 Emotional AI · 🔁 Metacognition · 📜 Persistent Memory · 🎙️ Voice Cloning
EXPLORATION · 06 Concept

Five Dimensions(5 LLMs) · In-depth

Personal, Relationship, Behavioral, Emotional & Historical is fine-tuned on a separate dimension of you. Here's exactly what each one knows, and how it speaks..

🔷 Five Dimensions · 🌀 Fusion Engine · 📊 Data → Digital · 💎 The Uniqueness · 💡 Why Blend? · 📖 Father's Light — Story · 🕊️ Father's Light — Legacy · ⚙️ Technical Build
EXPLORATION · 07 Simplifying

Talk to AI Like a Pro

10 techniques that unlock Claude's full potential.

📚 10 Techniques · 🎯 Hands-on Examples · ☁️ AWS & Cloud Examples · 🕹️ AI Playground
EXPLORATION · 08 Simplifying

How does an LLM actually answer you?

It never writes the full answer in one shot. It goes step by step — predicting a single next word, again and again, until a sentence appears. Here's the whole pipeline, one station at a time.

📝 Text · ✂️ Tokenization · 🔢 Token IDs · 🧮 Embeddings · 🧠 Transformer Layers · 📊 Logits · 📈 Softmax Probabilities · 🔮 Next Token · 🔁 Repeat · 💬 Answer
EXPLORATION · 09 Simplifying

How to build an LLM

The complete engineering path — data collection → processing → training → evaluation → deployment → continuous improvement — written the way a veteran would actually run the project. Every stage includes the models, tools, volumes, and hard-won decisions that separate a working behavioural LLM from an expensive experiment.

📍 Scope · 📥 Data Collection · 🏷️ Processing & Annotation · 🧠 Training · ✅ Evaluation · ☁️ Deployment · ♻️ Feedback Flywheel
EXPLORATION · 10 Concept

Hindi / Hinglish Voice Cloning Training Guide & Playground

Step-by-step process to create a perfect Indian-language voice model — covering data collection, phoneme mapping, training pipeline, and production deployment.

🎙️ Studio Recording · 🧬 Phoneme Mapping · ⚙️ Model Training · 🔁 Code-Switching · 🚀 Deployment
EXPLORATION · 11 Concept

Emotion Recognition

A CNN + LSTM hybrid architecture for Indian-language emotion analysis — spatial pattern extraction fused with sequential context, tuned for Hindi script and Hindi-English code-switching.

🧩 CNN Component • 🔗 LSTM Component • 🧠 Hybrid Fusion
EXPLORATION · 12 Concept

Every Color of Love

where technology meets the deepest human feeling.

🤝 91% Feel Accompanied by Their Twin · ❤️ 74% Hear Love in the Cloned Voice · 🌤️ Feel Less Lonely with AI Presence · ♾️ Moments of Love Preserved Forever
EXPLORATION · 13 Simplifying

Solving real problems

Technical systems and personal lives break for different reasons — but they respond to the same disciplined process. Six stages, applied through two lenses, turn "I don't know where to start" into a repeatable path forward.

💻 Technical Problems · 💗 Personal Problems
EXPLORATION · 14 Concept

LLM vs RAG vs RPA vs Agentic AI

Four terms that get used interchangeably in every boardroom — and shouldn't be. Each solves a different problem. Understanding the difference is the difference between picking the right tool and shipping an expensive mistake.

🧠 LLM • 🔍 RAG • 🤖 RPA • 🕹️ Agentic AI
EXPLORATION · 15 Concept

How LLMs actually generate text

You press Enter. Tokens stream back in milliseconds. Most people think that's magic. It's a pipeline. Here is every stage — from your keystroke to the typing effect on your screen.

✂️ Tokenizer · 🧮 Embeddings · 🧠 Transformer Blocks · 💾 KV Cache · 🎲 Sampling · ⚡ Speculative Decoding · 💬 Detokenize & Stream
Get in touch

If you care about
Human-Centric AI,
let's talk.

Open to advisory work, collaborations, and conversations on Emotional AI, Metacognition, Voice intelligence, and Human-Centric Intelligence.

Back to portfolio
Explorations Loading…
Loading exploration…