Ashutosh is available for hire
Hire AshutoshAshutosh Tripathi
Verified Expert in Engineering
Data Scientist and Developer
Gurugram, Haryana, India
Toptal member since September 14, 2021
Bio
Ashutosh is an AI architect and hands-on AI engineer specializing in machine learning, LLMs, and agentic AI. He builds production-ready AI systems that transform complex business problems into scalable products through architecture, rapid experimentation, and engineering. Ashutosh enjoys tackling ambiguous challenges where off-the-shelf solutions are insufficient, combining practical AI research with robust implementation to deliver measurable business impact.
Portfolio
LycaMobile
Artificial Intelligence (AI), AI Engineering, AI Architecture, Machine Learning...
vhAIre
Artificial Intelligence (AI), AI Engineering, AI Architecture...
Rakuten Symphony Singapore
Machine Learning, Data Science, Predictive Analytics, Predictive Modeling...
Experience
- Artificial Intelligence (AI) - 13 years
- Machine Learning - 11 years
- Large Language Models (LLMs) - 5 years
- Generative Artificial Intelligence (GenAI) - 5 years
- AI Architecture - 5 years
- Retrieval-augmented Generation (RAG) - 5 years
- Agentic AI - 4 years
- Model Context Protocol (MCP) - 2 years
Preferred Environment
Artificial Intelligence (AI), AI Engineering, Agentic AI, Machine Learning, Python, Large Language Models (LLMs), Retrieval-augmented Generation (RAG), Amazon Web Services (AWS), FastAPI
The most amazing...
...projects I've done involve building AI products that integrate machine learning, LLMs, agentic AI, analytics, and intelligent automation.
Work Experience
AI Architect
2025 - 2026
LycaMobile
- Architected an enterprise AI assistant enabling business users to query KPIs, perform RCA, analyze documents, and generate charts, CSVs, and presentations using natural language with LLMs, RAG, and agentic AI.
- Built an AI schema discovery engine that scans enterprise databases and data lakes, profiles datasets, and automatically generates LLM-ready semantic schemas, accelerating onboarding of new data sources for enterprise AI applications and AI agents.
- Designed production ML solutions for churn prediction and customer segmentation across 20+ countries, delivering €2+ million monthly business value with an agentic AI monitoring layer for model health, drift detection, and automated insights.
- Built a one-click AI competitive intelligence platform that automatically onboards competitors, extracts and structures product information using LLM-powered web intelligence, and generates comparison dashboards within minutes.
- Developed an AI email intelligence platform that lets users define complex monitoring and response rules in natural language, automatically prioritizing business-critical conversations and generating approval-based replies.
- Built an AI-powered network operations copilot that combines analytics, anomaly detection, root cause analysis, and LLM reasoning to accelerate incident investigation and operational decision-making.
- Created an autonomous AI code intelligence platform that uses AI agents to review enterprise SQL repositories, generate architecture and lineage diagrams, identify optimization opportunities, and provide conversational code understanding.
- Built an enterprise social intelligence platform that consolidates multi-channel social data, performs AI-powered sentiment and trend analysis, and enables conversational analytics using LLMs, RAG, and Text-to-SQL.
- Designed and implemented an enterprise model context protocol (MCP) platform enabling secure AI agent integration with enterprise tools through governance, guardrails, role-based access, auditing, and reusable connectors.
Technologies: Artificial Intelligence (AI), AI Engineering, AI Architecture, Machine Learning, Data Science, Deep Learning, Generative Artificial Intelligence (GenAI), Large Language Models (LLMs), OpenAI, Claude, Agentic AI, AI Agents, Prompt Engineering, Function Calling, Structured Outputs, Retrieval-augmented Generation (RAG), Semantic Search, Embeddings, Vector Search, Automation, AI Copilots, Conversational AI, Text-to-SQL, Natural Language Processing (NLP), LangChain, LangGraph, Pydantic, CrewAI, Claude Agent SDK, Python, FastAPI, Django, Flask, REST APIs, Asynchronous Programming, Amazon Web Services (AWS), SQL, Data Engineering, ETL, Analytics, Business Intelligence (BI), Machine Learning Operations (MLOps), Model Deployment, Model Monitoring, CI/CD Pipelines, Docker, Git, Linux, Artificial Neural Networks (ANN), Model Context Protocol (MCP), Claude Code, Amazon Bedrock AgentCore, AI Gateway, AI Agent Orchestration, Multi-agent Systems, AI Guardrails, Large Language Model Operations (LLMOps), Agent Memory, Tool Calling, Knowledge Graphs, Amazon Elastic Container Service (ECS), Amazon EC2, AWS Lambda, Amazon SageMaker, AWS ALB, Amazon Redshift, Amazon RDS, Amazon S3 (AWS S3), GitHub, GitHub Actions, Leadership, Project Management, Team Leadership, Technical Leadership, Product Management, Architecture
AI Architect
2024 - 2025
vhAIre
- Architected and led development of an AI-native recruitment platform that automates resume screening, candidate matching, AI interviews, and hiring workflows for recruiters and enterprises.
- Built an AI interviewer that conducts role-specific conversational interviews, evaluates technical and behavioral responses, and generates structured candidate summaries and fitment scores.
- Developed AI-powered resume parsing, job description generation, and semantic matching capabilities to improve candidate discovery and hiring efficiency.
- Designed an intelligent candidate recommendation engine combining LLMs, semantic search, and configurable scoring to connect candidates with relevant job opportunities.
- Led architecture and delivery of the end-to-end AI platform, guiding cross-functional engineering teams while remaining hands-on in designing and implementing core AI capabilities.
Technologies: Artificial Intelligence (AI), AI Engineering, AI Architecture, Large Language Models (LLMs), AI Agents, Agentic AI, Retrieval-augmented Generation (RAG), Semantic Search, Embeddings, Vector Search, Conversational AI, AI Interviewing, Resume Parsing, Candidate Matching, Intelligent Document Processing, Recommendation Systems, Machine Learning, Data Science, Deep Learning, Natural Language Processing (NLP), Classification, Algorithms, Information Extraction, Entity Recognition, Python, FastAPI, REST APIs, Microservices, Asynchronous Programming, OpenAI API, LangChain, LangGraph, Claude API, Hugging Face, Pydantic, Amazon Web Services (AWS), Docker, Kubernetes, PostgreSQL, Redis, SQL, Git, Enterprise Search, AI Workflow Automation, Knowledge Retrieval, SaaS, Recruitment Automation, Applicant Tracking Systems (ATS), HR Technology (HRtech), Resume Screening, Talent Intelligence, Talent Matching, Amazon S3 (AWS S3), Amazon Elastic Container Service (ECS), AWS Lambda, Amazon EC2, Amazon EKS, Amazon CloudFront, Amazon RDS, AWS IAM, GitHub Actions, CI/CD Pipelines, GitHub, Architecture, Leadership, Project Management, Startups, Team Leadership, Technical Leadership, Product Management
Principal Data Scientist (Part-time, Contract)
2024 - 2024
Rakuten Symphony Singapore
- Developed customer intelligence and segmentation models using behavioral, usage, and demographic data to enable targeted acquisition, upsell, and retention strategies.
- Built production machine learning models for churn prediction and customer propensity scoring, enabling proactive retention campaigns and data-driven customer engagement.
- Designed acquisition and growth analytics solutions that identified high-value prospects and optimized campaign prioritization through predictive modeling and customer analytics.
- Built an internal analytics copilot that enabled natural language exploration of business KPIs and customer metrics, reducing dependency on dashboards and manual SQL queries.
Technologies: Machine Learning, Data Science, Predictive Analytics, Predictive Modeling, Customer Analytics, Customer Segmentation, Churn Analysis, Propensity Modeling, Classification, Clustering, Feature Engineering, Statistical Modeling, Python, SQL, Scikit-learn, XGBoost, Pandas, NumPy, Data Analysis, Business Intelligence (BI), Data Visualization, ETL, Data Engineering, Amazon Web Services (AWS), Amazon Redshift, Telecommunications, Marketing Analytics, Customer Intelligence, CRM Analytics, GitHub Actions, GitHub
Senior Data Scientist
2021 - 2023
Asurion
- Built an AI-powered troubleshooting assistant using early RAG architecture and OpenAI models, enabling support teams to resolve device issues through enterprise knowledge retrieval and conversational assistance.
- Developed recommendation systems serving over 2 million users, personalizing content and support journeys to improve customer engagement and self-service experiences.
- Designed semantic search solutions using fine-tuned sentence embeddings and hybrid retrieval techniques to improve FAQ discovery, chatbot accuracy, and customer support automation.
- Built computer vision models for automated device damage assessment and AI-driven fraud detection, enabling consistent damage classification and improved insurance claim validation.
Technologies: Machine Learning, Computer Vision, Natural Language Processing (NLP), TensorFlow, PyTorch, Data Science, Deep Learning, Recommendation Systems, Semantic Search, Information Retrieval, Retrieval-augmented Generation (RAG), OpenAI API, Prompt Engineering, Keras, Hugging Face, Scikit-learn, EfficientNet, Universal Sentence Encoder, Sentence Embeddings, Semantic Similarity, Text Classification, Intent Classification, Hybrid Search, Tf-idf, Word Mover’s Distance (WMD), Artificial Neural Networks (ANN), Amazon RDS, Amazon Elastic Container Service (ECS), Amazon S3 (AWS S3), Amazon EC2, AWS Lambda, Amazon Elastic MapReduce (EMR), GitHub, GitHub Actions
AI Consultant (Part-time)
2020 - 2022
Cybo
- Designed and evaluated custom 3D CNN architectures by extending VGG, ResNet, and EfficientNet models for automated classification of gigapixel whole-slide digital cytology images.
- Developed a multiple instance learning (MIL) framework that leveraged slide-level annotations to efficiently train deep learning models by prioritizing high-confidence image patches during backpropagation.
- Designed a novel patch-embedding aggregation pipeline that transformed whole-slide image embeddings into compact feature maps, enabling efficient CNN-based classification while preserving critical cellular characteristics.
Technologies: Artificial Intelligence (AI), Deep Learning, Computer Vision, Healthcare, Digital Cytology, Medical Imaging, Whole Slide Imaging (WSI), Multiple Instance Learning (MIL), Weakly Supervised Learning, Image Classification, Convolutional Neural Networks (CNNs), 3D CNN, Residual Neural Networks (ResNets), EfficientNet, Transfer Learning, TensorFlow, Keras, PyTorch, Applied AI Research, Model Architecture Design, Feature Engineering, Model Tuning, Scientific Computing, Python, OpenCV
Data Scientist (Part-time)
2021 - 2021
Relativ
- Partnered with strategy teams to analyze consumer and market research data, delivering actionable insights that informed brand positioning and growth strategies for global consumer brands.
- Applied statistical analysis and hypothesis testing to validate business assumptions, quantify consumer behavior, and support evidence-based strategic decision-making.
- Designed analytical studies and executive-ready reports that translated complex data into clear business recommendations for cross-functional strategy and leadership teams.
Technologies: Statistical Analysis, Hypothesis Testing, Experimental Design, A/B Testing, Statistical Modeling, Data Science, Business Analytics, Marketing Analytics, Market Research, Exploratory Data Analysis, Python, SQL, Pandas, NumPy, Decision Support, Strategy Analytics, Executive Reporting
Data Scientist
2019 - 2020
Rakuten Mobile
- Developed and deployed production NLP models for Japanese customer interactions using hybrid deep learning architectures, enabling sentiment analysis, topic classification, and risk scoring across omnichannel support platforms.
- Built AI-driven network capacity forecasting solutions by combining time-series forecasting, regression models, and telecom domain intelligence to proactively predict cell capacity and optimize network planning.
- Designed time-series anomaly detection models for telecom network KPIs using statistical and machine learning techniques, enabling proactive identification of abnormal network behavior.
- Productionized machine learning solutions using Flask, Docker, and Kubernetes, enabling scalable deployment of AI models into enterprise telecom applications.
Technologies: Natural Language Processing (NLP), Docker, Kubernetes, Deep Learning, Flask, Python, Machine Learning, Time Series Forecasting, Anomaly Detection, Regression, Classification, Sentiment Analysis, Topic Modeling, LSTM, Attention Mechanisms, ARIMA, Holt-Winters, REST APIs, Artificial Neural Networks (ANN)
Data Scientist
2017 - 2019
Samsung Research
- Built and deployed a machine learning solution that automatically classified QA-reported network issues and routed them to the appropriate engineering teams, reducing manual triage and improving issue resolution efficiency.
- Developed an Android application implementing 3GPP SIM file system parsing, enabling QA engineers to inspect and modify SIM data directly from mobile devices, replacing complex desktop-based testing workflows.
- Developed machine learning and statistical analytics solutions to process large-scale Android and telecom protocol logs, enabling anomaly detection and faster root cause analysis of network and device issues.
Technologies: Machine Learning, Long-term Evolution (LTE), Python, SQL, Statistical Modeling, Classification, Anomaly Detection, Predictive Analytics, Flask, XGBoost, Log Analytics, Feature Engineering, Data Mining, Data Processing, Mobile Networks, Android, Root-cause Analysis (RCA), Network Diagnostics, Protocol Analysis, Artificial Neural Networks (ANN)
Experience
Enterprise Decision Intelligence Platform
Designed and led the development of an enterprise AI platform that enables business and technical teams to interact with enterprise data and knowledge through natural language. Instead of relying on traditional dashboards or manual SQL, users can retrieve KPIs, perform root cause analysis, generate executive-ready PowerPoint presentations, create visualizations, export datasets, and analyze uploaded documents from a unified AI interface.
The platform was built using a modular, metadata-driven architecture that supports rapid onboarding of new business domains, data sources, and analytical workflows without modifying the core orchestration framework. It combines LLMs, hybrid retrieval-augmented generation (RAG), Text-to-SQL, agentic AI patterns, workflow orchestration, memory, and reusable tool integrations to deliver secure, context-aware responses across structured and unstructured enterprise data.
As the AI platform architect, I owned the overall architecture, technology strategy, AI design patterns, engineering standards, and core implementation while mentoring engineers and driving the platform’s long-term technical evolution.
The platform was built using a modular, metadata-driven architecture that supports rapid onboarding of new business domains, data sources, and analytical workflows without modifying the core orchestration framework. It combines LLMs, hybrid retrieval-augmented generation (RAG), Text-to-SQL, agentic AI patterns, workflow orchestration, memory, and reusable tool integrations to deliver secure, context-aware responses across structured and unstructured enterprise data.
As the AI platform architect, I owned the overall architecture, technology strategy, AI design patterns, engineering standards, and core implementation while mentoring engineers and driving the platform’s long-term technical evolution.
Adaptive AI Hiring & Interview Platform
https://www.vhaire.com/Designed and led the development of an AI-native recruitment platform that assists organizations throughout the complete hiring lifecycle, from candidate sourcing to final interview evaluation. The platform automatically parses resumes and job descriptions, extracts structured skills and experience, generates candidate-job matching scores, recommends suitable applicants, and conducts adaptive AI interviews that dynamically adjust question difficulty based on candidate responses.
The interview engine combines LLMs, retrieval-augmented generation (RAG), structured evaluation frameworks, resume understanding, and domain-specific knowledge to generate personalized technical and behavioral interviews while producing detailed competency reports and hiring recommendations. Recruiters can review AI-generated scorecards, interview transcripts, and evidence-based evaluations before making final decisions.
As an AI architect, I led overall product architecture, AI strategy, engineering direction, and core implementation, while managing a multidisciplinary engineering team responsible for AI, back-end, front-end, and platform development.
The interview engine combines LLMs, retrieval-augmented generation (RAG), structured evaluation frameworks, resume understanding, and domain-specific knowledge to generate personalized technical and behavioral interviews while producing detailed competency reports and hiring recommendations. Recruiters can review AI-generated scorecards, interview transcripts, and evidence-based evaluations before making final decisions.
As an AI architect, I led overall product architecture, AI strategy, engineering direction, and core implementation, while managing a multidisciplinary engineering team responsible for AI, back-end, front-end, and platform development.
Enterprise AI Integration Platform (MCP)
Designed the next-generation AI integration platform for enterprise AI applications by adopting the model context protocol (MCP) as a standardized interface between AI agents and enterprise systems. The goal was to eliminate duplicated integrations across multiple AI products by centralizing enterprise tools, prompts, governance, and reusable capabilities into a unified platform.
The architecture exposes databases, cloud storage, APIs, and internal services as reusable MCP tools, allowing AI applications to focus on agent orchestration while the platform manages secure tool execution and governance. Built on AWS ECS using FastMCP wrapped with FastAPI, it preserves the MCP JSON-RPC standard while enabling enterprise-grade deployment. The design incorporates reusable connectors, RBAC, tool permissions, SQL guardrails, audit logging, and standardized execution pipelines to securely integrate AI agents with Redshift, S3, GitHub, REST APIs, and internal enterprise systems.
As an AI platform architect, I defined the reference architecture, integration strategy, and engineering standards driving Lyca’s transition toward a reusable enterprise AI ecosystem.
The architecture exposes databases, cloud storage, APIs, and internal services as reusable MCP tools, allowing AI applications to focus on agent orchestration while the platform manages secure tool execution and governance. Built on AWS ECS using FastMCP wrapped with FastAPI, it preserves the MCP JSON-RPC standard while enabling enterprise-grade deployment. The design incorporates reusable connectors, RBAC, tool permissions, SQL guardrails, audit logging, and standardized execution pipelines to securely integrate AI agents with Redshift, S3, GitHub, REST APIs, and internal enterprise systems.
As an AI platform architect, I defined the reference architecture, integration strategy, and engineering standards driving Lyca’s transition toward a reusable enterprise AI ecosystem.
AI-powered Schema Discovery Platform
Designed and built an AI-powered schema discovery platform that automatically transforms enterprise databases into AI-ready knowledge layers. The platform connects to relational databases, cloud storage, and enterprise data sources to discover schemas, profile datasets, compute descriptive statistics, and generate business-friendly metadata without requiring manual documentation.
Instead of simply documenting tables, the platform uses LLMs to infer business meaning, relationships, KPIs, and semantic context, producing structured metadata optimized for downstream AI systems such as Text-to-SQL, retrieval-augmented generation (RAG), and agentic workflows. This significantly reduces the effort required to onboard new business domains while improving the accuracy and reliability of AI-generated responses.
The platform follows a connector-based architecture supporting extensible integrations, automated metadata generation, AI-ready semantic schemas, and reusable ingestion pipelines, enabling rapid adoption across heterogeneous enterprise environments.
Instead of simply documenting tables, the platform uses LLMs to infer business meaning, relationships, KPIs, and semantic context, producing structured metadata optimized for downstream AI systems such as Text-to-SQL, retrieval-augmented generation (RAG), and agentic workflows. This significantly reduces the effort required to onboard new business domains while improving the accuracy and reliability of AI-generated responses.
The platform follows a connector-based architecture supporting extensible integrations, automated metadata generation, AI-ready semantic schemas, and reusable ingestion pipelines, enabling rapid adoption across heterogeneous enterprise environments.
AI-powered GitHub Code Review Platform
Designed and led the development of an AI-powered code intelligence platform that automatically reviews GitHub repositories and transforms complex codebases into business and engineering insights. Instead of producing generic code summaries, the platform analyzes architecture, business logic, dependencies, SQL procedures, APIs, and application workflows to generate executive-friendly explanations alongside deep technical documentation.
The system combines LLMs, retrieval-augmented generation (RAG), multi-agent orchestration, and repository analysis to produce architecture diagrams, Mermaid visualizations, data lineage, dependency graphs, code quality findings, performance optimization recommendations, security observations, and conversational Q&A over the analyzed codebase. It supports multiple programming languages and enables both technical engineers and business stakeholders to understand unfamiliar systems within minutes rather than days.
As an AI architect, I designed the overall platform architecture, agent workflows, prompt strategy, review pipeline, and AI orchestration, and led implementation of the core intelligence engine.
The system combines LLMs, retrieval-augmented generation (RAG), multi-agent orchestration, and repository analysis to produce architecture diagrams, Mermaid visualizations, data lineage, dependency graphs, code quality findings, performance optimization recommendations, security observations, and conversational Q&A over the analyzed codebase. It supports multiple programming languages and enables both technical engineers and business stakeholders to understand unfamiliar systems within minutes rather than days.
As an AI architect, I designed the overall platform architecture, agent workflows, prompt strategy, review pipeline, and AI orchestration, and led implementation of the core intelligence engine.
AI for Whole-slide Digital Cytology
Researched and developed deep learning techniques for automated analysis of gigapixel whole-slide digital cytology images used for detecting cervical cellular abnormalities, including NILM, LSIL, HSIL, and Adenocarcinoma. The primary challenge was enabling efficient learning from extremely large whole-slide images where detailed cell-level annotations were scarce and conventional CNN architectures could not be applied directly.
Designed custom 3D CNN architectures by extending VGG, ResNet, and EfficientNet for volumetric cytology analysis, and developed a multiple instance learning (MIL) training framework that leveraged slide-level labels by selecting high-confidence image patches for backpropagation. Additionally, designed a novel patch-embedding aggregation pipeline that transformed whole-slide representations into compact feature maps, enabling efficient CNN-based classification while preserving diagnostically important cellular characteristics.
The project involved extensive experimentation, research, and model optimization using TensorFlow, Keras, and PyTorch to advance AI techniques for digital cytology.
Designed custom 3D CNN architectures by extending VGG, ResNet, and EfficientNet for volumetric cytology analysis, and developed a multiple instance learning (MIL) training framework that leveraged slide-level labels by selecting high-confidence image patches for backpropagation. Additionally, designed a novel patch-embedding aggregation pipeline that transformed whole-slide representations into compact feature maps, enabling efficient CNN-based classification while preserving diagnostically important cellular characteristics.
The project involved extensive experimentation, research, and model optimization using TensorFlow, Keras, and PyTorch to advance AI techniques for digital cytology.
AI-powered UX Testing Platform
Designed and developed a multimodal AI platform that automates usability testing directly from Figma prototypes. Instead of requiring human usability sessions, the platform analyzes Figma designs, generates realistic user personas, creates task-specific test scenarios, simulates user journeys, identifies UX friction, and produces structured usability reports with actionable recommendations.
The core engineering challenge was transforming extremely large and complex Figma JSON structures into concise, AI-ready representations that preserved design hierarchy, navigation flows, components, and interaction context while remaining efficient for LLM reasoning. The platform orchestrates a sequential AI pipeline covering design interpretation, persona generation, scenario creation, usability simulations, UX synthesis, and automated report generation.
Designed as an extensible architecture, the platform can evolve to incorporate multimodal vision models, enabling AI to reason directly over rendered interfaces in addition to structured design metadata.
The core engineering challenge was transforming extremely large and complex Figma JSON structures into concise, AI-ready representations that preserved design hierarchy, navigation flows, components, and interaction context while remaining efficient for LLM reasoning. The platform orchestrates a sequential AI pipeline covering design interpretation, persona generation, scenario creation, usability simulations, UX synthesis, and automated report generation.
Designed as an extensible architecture, the platform can evolve to incorporate multimodal vision models, enabling AI to reason directly over rendered interfaces in addition to structured design metadata.
Telecom Bundle Portfolio Optimization using Integer Linear Programming
Designed and developed an optimization engine to recommend the most cost-effective wholesale bundle portfolio for a global MVNO by formulating the Retail Minus (RM) allocation problem as an Integer Linear Programming (ILP) model. Instead of relying on manual commercial planning, the solution mathematically optimized bundle allocation across customer segments while satisfying complex business, pricing, and eligibility constraints.
The optimization model minimized overall bundle procurement cost while ensuring complete customer demand coverage through optimal selection of eligible bundle combinations. Commercial rules, bundle validity, customer lifecycle stages, renewal policies, allocation limits, and product-specific eligibility were translated into mathematical constraints, enabling thousands of allocation decisions to be solved automatically for every optimization cycle. Built using Python, Pandas, and PuLP with the CBC solver, the platform generated optimized bundle recommendations, cost comparisons, utilization analysis, overage and underage reporting, and executive dashboards supporting Pricing, Finance, and Wholesale teams in commercial decision-making.
The optimization model minimized overall bundle procurement cost while ensuring complete customer demand coverage through optimal selection of eligible bundle combinations. Commercial rules, bundle validity, customer lifecycle stages, renewal policies, allocation limits, and product-specific eligibility were translated into mathematical constraints, enabling thousands of allocation decisions to be solved automatically for every optimization cycle. Built using Python, Pandas, and PuLP with the CBC solver, the platform generated optimized bundle recommendations, cost comparisons, utilization analysis, overage and underage reporting, and executive dashboards supporting Pricing, Finance, and Wholesale teams in commercial decision-making.
AI-powered Competitive Pricing Intelligence Platform
Designed and developed an AI-powered competitive pricing intelligence platform that continuously monitors competitor products, bundles, promotions, and pricing strategies to support commercial decision-making. Instead of relying on manual market research, business users define competitors and extraction requirements using natural language, allowing the platform to automatically collect, structure, compare, and analyze competitor offerings.
The platform uses Firecrawl to efficiently discover and scrape relevant competitor pages while optimizing API costs through a multi-stage pipeline. Website maps are first analyzed to identify pricing-related pages before selectively extracting content, which is then structured and validated using LLMs. For broader use cases beyond telecom, the platform can switch to Firecrawl’s Agent API for autonomous web exploration. Results are presented through interactive comparison dashboards with structured plan catalogs, side-by-side competitor benchmarking, executive summaries, and conversational AI that allows users to query the latest collected market intelligence using natural language.
The platform uses Firecrawl to efficiently discover and scrape relevant competitor pages while optimizing API costs through a multi-stage pipeline. Website maps are first analyzed to identify pricing-related pages before selectively extracting content, which is then structured and validated using LLMs. For broader use cases beyond telecom, the platform can switch to Firecrawl’s Agent API for autonomous web exploration. Results are presented through interactive comparison dashboards with structured plan catalogs, side-by-side competitor benchmarking, executive summaries, and conversational AI that allows users to query the latest collected market intelligence using natural language.
AI-powered Email Intelligence Assistant
An AI-powered email intelligence platform that helps users manage critical business communications through natural language rather than traditional rule-based filters. Instead of configuring complex email rules, users describe what matters to them using free text, such as important people, business situations, approvals, or follow-up expectations, and the platform continuously adapts its monitoring and prioritization accordingly.
The solution integrates with Microsoft Graph and Outlook to analyze incoming emails using LLMs, enterprise context, and user-defined intent. AI evaluates sender hierarchy, organizational relationships, email content, urgency, deadlines, approvals, and business context to determine importance, trigger reminders, recommend responses, and identify emails requiring immediate attention. Users receive intelligent notifications through Microsoft Teams, manage personalized monitoring rules through a web dashboard, and can subscribe or unsubscribe from conversation threads at any time.
Unlike conventional email rules or AI summarization tools, the platform continuously interprets evolving business intent using natural language, enabling personalized enterprise email intelligence rather than static automation.
The solution integrates with Microsoft Graph and Outlook to analyze incoming emails using LLMs, enterprise context, and user-defined intent. AI evaluates sender hierarchy, organizational relationships, email content, urgency, deadlines, approvals, and business context to determine importance, trigger reminders, recommend responses, and identify emails requiring immediate attention. Users receive intelligent notifications through Microsoft Teams, manage personalized monitoring rules through a web dashboard, and can subscribe or unsubscribe from conversation threads at any time.
Unlike conventional email rules or AI summarization tools, the platform continuously interprets evolving business intent using natural language, enabling personalized enterprise email intelligence rather than static automation.
AI-powered Network Capacity Planning Platform
Designed and developed an AI-powered network capacity planning platform to help network planning teams proactively forecast LTE cell capacity requirements across Japan. Rather than forecasting numerous radio KPIs independently, the platform adopted a hybrid machine learning approach by identifying active users and data traffic volume as the most stable long-term predictors and estimating future PRB utilization through regression modeling. The predicted values were then combined with standardized capacity formulas to estimate future network capacity over six- and twelve-month planning horizons.
The platform uses an ensemble forecasting framework where multiple models, including ARIMA, Prophet, Holt-Winters, and LSTM, are evaluated independently for every network cell, with the best-performing model automatically selected based on historical validation results. Monthly forecasts are presented through interactive dashboards, enabling network planning and performance management teams to identify future capacity constraints and prioritize infrastructure expansion using data-driven insights instead of reactive planning.
The platform uses an ensemble forecasting framework where multiple models, including ARIMA, Prophet, Holt-Winters, and LSTM, are evaluated independently for every network cell, with the best-performing model automatically selected based on historical validation results. Monthly forecasts are presented through interactive dashboards, enabling network planning and performance management teams to identify future capacity constraints and prioritize infrastructure expansion using data-driven insights instead of reactive planning.
Enterprise Semantic Knowledge Retrieval Platform
Designed and developed a large-scale semantic knowledge retrieval platform powering customer self-service support for insured mobile devices within NTT Docomo’s customer application. The solution enabled users to diagnose issues such as network connectivity, application crashes, Bluetooth failures, and screen defects by retrieving the most relevant troubleshooting articles from thousands of continuously evolving knowledge base documents.
Instead of relying solely on keyword search, the platform implemented a hybrid retrieval engine combining multilingual Universal Sentence Encoder embeddings with cosine similarity, Word Mover’s Distance (WMD), TF-IDF similarity, exact keyword matching, and statistical ranking techniques. Individual scores were normalized and combined through a weighted ensemble, where optimal weights were learned using grid-search optimization against a curated golden evaluation dataset. The platform also automated FAQ version synchronization to ensure users always received the latest troubleshooting guidance while maintaining low-latency semantic search at production scale.
Instead of relying solely on keyword search, the platform implemented a hybrid retrieval engine combining multilingual Universal Sentence Encoder embeddings with cosine similarity, Word Mover’s Distance (WMD), TF-IDF similarity, exact keyword matching, and statistical ranking techniques. Individual scores were normalized and combined through a weighted ensemble, where optimal weights were learned using grid-search optimization against a curated golden evaluation dataset. The platform also automated FAQ version synchronization to ensure users always received the latest troubleshooting guidance while maintaining low-latency semantic search at production scale.
AI-powered Customer Churn Intelligence Platform
Designed and developed an enterprise-scale customer churn intelligence platform to proactively identify subscribers likely to leave within the next 60 days across Lyca Mobile’s global MVNO operations. Supporting more than 20 countries, including the UK, France, the USA, Germany, Italy, and Belgium, the platform enables Customer Value Management (CVM) teams to launch targeted retention campaigns before customers disengage.
The solution combines large-scale feature engineering across customer usage, recharge behavior, network KPIs, demographics, complaints, roaming activity, device characteristics, and other behavioral signals. Multiple machine learning models, including XGBoost, CatBoost, LightGBM, and Random Forest, were evaluated independently for each country, allowing the best-performing model to be automatically selected based on local customer behavior. The platform automates data preparation, model training, calibration, scoring, risk segmentation, monitoring, and dashboard generation, producing monthly churn probabilities that directly support personalized retention initiatives. A key focus of the project was continuous feature experimentation and country-specific optimization rather than relying on a single global model.
The solution combines large-scale feature engineering across customer usage, recharge behavior, network KPIs, demographics, complaints, roaming activity, device characteristics, and other behavioral signals. Multiple machine learning models, including XGBoost, CatBoost, LightGBM, and Random Forest, were evaluated independently for each country, allowing the best-performing model to be automatically selected based on local customer behavior. The platform automates data preparation, model training, calibration, scoring, risk segmentation, monitoring, and dashboard generation, producing monthly churn probabilities that directly support personalized retention initiatives. A key focus of the project was continuous feature experimentation and country-specific optimization rather than relying on a single global model.
Education
2020 - 2021
Post Graduation Program in AI-ML in Artificial Intelligence
McCombs School of Business, University of Texas at Austin - Remote
2013 - 2017
Bachelor's Degree in Computer Science
Indian Institute of Technology, Patna - Patna, India
Certifications
FEBRUARY 2021 - PRESENT
AI-ML Post Graduate Certification
McCombs School of Business, University of Texas at Austin
DECEMBER 2020 - PRESENT
The Introduction to Quantum Computing
Saint Petersburg State University
JULY 2020 - PRESENT
Docker and Kubernetes: The Complete Guide
Udemy
MAY 2020 - PRESENT
Taming Big Data with Apache Spark and Python — Hands On!
Udemy
AUGUST 2019 - PRESENT
Python for Time Series Data Analysis
Udemy
Skills
Libraries/APIs
REST APIs, TensorFlow, Keras, Scikit-learn, LSTM, SpaCy, PyTorch, Pandas, Pydantic, OpenAI API, Claude API, XGBoost, NumPy, JSON-RPC, CatBoost, OpenCV
Tools
ARIMA, Claude, Text-to-SQL, Claude Agent SDK, Git, Claude Code, GitHub, Amazon Elastic Container Service (ECS), Amazon SageMaker, AWS IAM, Amazon Elastic MapReduce (EMR), Mermaid, Figma, Amazon EKS, Amazon CloudFront
Languages
Python, SQL, Java, C++
Frameworks
Spark, Django, Flask, LangGraph, FastMCP, LightGBM, Apache Spark
Paradigms
Automation, Asynchronous Programming, Business Intelligence (BI), Microservices, Anomaly Detection, Model Context Protocol (MCP), Role-based Access Control (RBAC), Mathematical Optimization, Distributed Computing, ETL, Human-computer Interaction (HCI), Constraint Programming
Platforms
Docker, Amazon Web Services (AWS), CrewAI, Linux, Amazon EC2, AWS Lambda, AWS ALB, Kubernetes, Android
Storage
PostgreSQL, Databases, JSON, Amazon S3 (AWS S3), MongoDB, Redis
Industry Expertise
Project Management, Telecommunications, Healthcare
Other
Natural Language Processing (NLP), Machine Learning, Statistics, Algorithms, Deep Learning, Artificial Intelligence (AI), Recommendation Systems, Big Data, Forecasting, Regression, Mathematics, Sentiment Analysis, Machine Translation, Artificial Neural Networks (ANN), Convolutional Neural Networks (CNNs), Generative Pre-trained Transformers (GPT), Data Science, AI Engineering, AI Architecture, Generative Artificial Intelligence (GenAI), Agentic AI, AI Agents, Multi-agent Systems, Semantic Search, Vector Search, Large Language Models (LLMs), Retrieval-augmented Generation (RAG), FastAPI, Data Structures, OpenAI, Prompt Engineering, Function Calling, Structured Outputs, Embeddings, AI Copilots, Conversational AI, LangChain, Data Engineering, Analytics, Machine Learning Operations (MLOps), Model Deployment, Model Monitoring, CI/CD Pipelines, AI Interviewing, Intelligent Document Processing, Classification, Information Extraction, Entity Recognition, Hugging Face, Enterprise Search, AI Workflow Automation, Knowledge Retrieval, SaaS, Predictive Analytics, Predictive Modeling, Customer Analytics, Customer Segmentation, Churn Analysis, Propensity Modeling, Clustering, Feature Engineering, Statistical Modeling, Data Analysis, Customer Intelligence, Information Retrieval, EfficientNet, Universal Sentence Encoder, Sentence Embeddings, Semantic Similarity, Text Classification, Intent Classification, Hybrid Search, Tf-idf, Topic Modeling, Attention Mechanisms, Holt-Winters, Data Processing, Root-cause Analysis (RCA), Image Classification, 3D CNN, Residual Neural Networks (ResNets), Transfer Learning, Model Architecture Design, Model Tuning, Statistical Analysis, Hypothesis Testing, Experimental Design, Business Analytics, Market Research, Exploratory Data Analysis, Decision Support, Strategy Analytics, Executive Reporting, Amazon Bedrock AgentCore, AI Gateway, AI Agent Orchestration, AI Guardrails, Large Language Model Operations (LLMOps), Agent Memory, Tool Calling, Software Architecture, System Design, Metadata Management, Semantics, Enterprise AI, Workflow Orchestration, Enterprise Integration, AWS ECS Fargate, Load Balancers, API Design, Competitive Intelligence, Price Analysis, Firecrawl, Web Scraping, Data Structuring, Workflow Automation, Notification Systems, Ensemble Learning, Dashboard Development, Capacity Planning, Random Forests, Word Mover’s Distance, Cosine Similarity, Search Ranking, Amazon RDS, GitHub Actions, Linear Algebra, Leadership, Team Leadership, Technical Leadership, Product Management, Architecture, Startups, Computer Vision, Time Series Analysis, Time Series, Long-term Evolution (LTE), 5G, Quantum Computing, Operating Systems, Deep Reinforcement Learning, Resume Parsing, Candidate Matching, Recruitment Automation, Applicant Tracking Systems (ATS), HR Technology (HRtech), Resume Screening, Talent Intelligence, Talent Matching, Data Visualization, Amazon Redshift, Marketing Analytics, CRM Analytics, Word Mover’s Distance (WMD), Time Series Forecasting, Log Analytics, Data Mining, Mobile Networks, Network Diagnostics, Protocol Analysis, Digital Cytology, Medical Imaging, Whole Slide Imaging (WSI), Multiple Instance Learning (MIL), Weakly Supervised Learning, Applied AI Research, Scientific Computing, A/B Testing, Knowledge Graphs, UX Research, Product Design, Design Systems, Microsoft Graph API, Integer Linear Programming, Operations Research, PuLP, CBC Solver, Optimization Modeling, Pricing Strategy, Revenue Optimization, Calculus, Quantum Physics, Quantum Mechanics
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring