Work

A chronological timeline of my career milestones and projects

O.XYZ

Lead AI Engineer
Sep 2024 - Present

OriginStudio - Vibe Coding Tool

Web based Vibe Coding tool that lets you create production ready projects with FE/BE from scratch with MCP integration. Like Cursor, v0 and Lovable

Category: Generative AI
Domain: Code Generation
Technologies:
Coding Agents MCP Sandbox Gitea Supabase Logto

OCEAN - AI Search Engine

One of the world's fastest AI Search engine, powered by Cerebras chips, to take on the likes of Perplexity; supports deep research, document upload, tool calling, agentic behavior, public APIs

Category: Generative AI
Domain: AI Search
Technologies:
LLMs Guardrails LangGraph MCP FastAPI Cerebras RAG

Miss O - Voice Assistant

Low Latency Voice Assistant for voice to voice conversation with AI; also developed a version for underage audience designed with utmost ethical considerations

Category: Generative AI
Domain: Voice Assistant
Technologies:
VAD STT (Inhouse) RAG LLMs TTS (Cerebras, ElevenLabs) LiveKit

Voice Authentication

Authenticate user to login to the O.XYZ systems via their Voice, based on their latent audio biometric details

Category: Generative AI
Domain: Authentication
Technologies:
Audio AI Models Semantic match Privy integration FastAPI Security

O Routing Intelligence

A dynamic framework that utilizes multiple LLMs to improve task-specific accuracy and efficiency by intelligently routing queries to the most suitable model; Improved performance on MMLU, MuSR, ARC, and BBH

Category: Generative AI
Domain: LLM Optimization
Technologies:
Routing Intelligence Constraint optimization Query to Task identification

Outplay

Lead Data Scientist
Nov 2021 - Aug 2024

SureConnect.ai

AI powered phone number validating product that 10x the dial-to-connect ratio. Automatically calls and converses with prospect to identify if it has reached correct number or not, find the best to call, detect call outcome

Category: Product
Domain: Sales Intelligence
Technologies:
Speech Recognition Text-to-Speech LLM Dialog Management Twilio

Conversation Intelligence

AI powered meeting analysis product that joins ongoing calls to identify prospect's concerns, suggesting answers, perform live mood detection and after the call ends, performs further advanced analysis

Category: Product
Domain: Sales Intelligence
Technologies:
Speech Recognition Speaker Identification Mentions Identification

SalesEmail Assistant

Multi-task AI assistant that writes highly personalized mail from scratch for sales representatives and provides recommendations to enhance existing sales mails by gamifying the complete process

Category: Product
Domain: Sales Intelligence
Technologies:
Mail Scoring Sentiment Tones Spam words Paraphrasing

In-House Transcription System

Speech-to-text (ASR) services for live and offline sales call transcription; comparable to the likes of AWS Transcribe

Category: Audio Intelligence
Domain: Speech Recognition
Technologies:
Wav2Vec2 HuBERT WordBoost Custom Vocabulary Websocket

Speaker Diarization and Identification

Segregate and recognize different speakers in a call by comparing against the available speaker's voice prints

Category: Audio Intelligence
Domain: Speech Recognition
Technologies:
SpeechBrain Bi-encoder Audio Embeddings

SalesGPT

Training and Deploying inhouse LLMs to make it more domain aware for Sales related downstream tasks. Explored 3rd party models like GPT-4 for evaluation

Category: Generative AI
Domain: LLM Training
Technologies:
GPT4 ChatGPT LLaMA Quantization QLoRA Flash Attention LangChain

Voice Cloning

Inhouse models to support creation of fast and accurate voice clones of Non-Western voices with very few examples or with fine-tuning approach

Category: Generative AI
Domain: Audio Synthesis
Technologies:
XTTSv2 tortoise-tts Transformers Quantization LoRA

AlgoAnalytics

Senior Data Scientist
Mar 2021 - Nov 2021

Indian Companies Knowledge Graph

Create a KG containing the enterprise-level details for Indian companies (like Crunchbase); analyze graph to find similar company and recommend potential acquisitions and investments

Category: NLP
Domain: Knowledge Graphs
Technologies:
Knowledge Graphs Recommendation System Information Extraction

Social Media Trend Analyzer

Analyze the trend of product, person or any entity on Social media like Twitter

Category: NLP
Domain: Social Media Analytics
Technologies:
NER Social media scraper Trend forecasting Sentiment Tones

TCS

Data Scientist
Aug 2016 - Feb 2021

Resolution Notes Mining

Mine L1/L2/L3's alert resolution notes to generate rules to automate frequently occurring errors

Category: NLP
Domain: IT Operations
Technologies:
Pattern finding and matching Association rule mining

Enterprise Revenue Prediction

Revenue forecasting of company's hierarchical business units with ~91% accuracy

Category: Forecasting
Domain: Business Analytics
Technologies:
Time series forecasting ARIMA Holt Winters HTS

Enterprise Sales

Hybrid recommendation system to generate B2B suggestions for company's Sales team with major focus on accuracy and explainability

Category: Recommendation System
Domain: Sales Analytics
Technologies:
Recommendation Systems Cross-Sell Up-Sell Day 0 Sell

System Expert Assignment

Given an IT system alert, automatically find the best L1/L2/L3 resolver and assign the task; done by performing multiple objective optimization, leading to reduction of overall MTTR by 30%

Category: Optimization
Domain: IT Operations
Technologies:
Ranking algorithms Multi-objective optimization Resource allocation

Personalized Pricing Model

Find optimal Batch analytics pricing by minimizing jobs in analysis with minimum adverse effect on SLA prediction accuracy

Category: Optimization
Domain: Pricing Strategy
Technologies:
Graph reduction Subset finding Price optimization

Root-Cause Analysis

Incident management on structural graph of the enterprise to identify correct root cause for a system level event

Category: Causality Analysis
Domain: IT Operations
Technologies:
Correlation vs Causality Graph traversal Multi-modal analysis

What-If Analysis

Model causality graph of an enterprise to perform intervention on behavior of services and ask counterfactual business analysis questions

Category: Causality Analysis
Domain: Business Analytics
Technologies:
Product Behavior Simulation Confounding analysis Prediction

Business Card Reader

OCR engine to extract entities from multilingual business cards for ultra-fast personnel contact onboarding

Category: Computer Vision
Domain: OCR
Technologies:
Optical Character Recognition NER Validity check

Math Formula Recognition

Extract handwritten math expressions from documents for fast digitalization and education purposes

Category: Computer Vision
Domain: OCR
Technologies:
Optical Character Recognition LaTeX n-gram model

FAQ Chatbot

Virtual assistant for project's flagship solution that increased daily impressions by 10% and reduced human intervention by 40%

Category: NLP
Domain: Customer Service
Technologies:
Chatbot Question and Answer identification

C-DAC

Research Intern
2015

IoT Speech Recognition Engine

Design and developed IoT Speech Recognition engine in Indian languages and deployed it on low-power hardware as a low-cost solution for the rural population

Category: Speech Recognition
Domain: IoT
Technologies:
Speech Recognition IoT Indian languages