~/chicago $ whoami

I build agentic AI systems

AI/ML Engineer with 7+ years across the full ML lifecycle — conversational & voice AI, LLMs, agentic systems and computer vision, from architecture to production monitoring.

See my work Get in touch
Portrait of Umme Athiya
▲ LLMs · RAG · Agents
● Conversational AI @ Aetna
◆ ex-IBM · DePaul MS AI
hover to fire neurons · click for a forward pass
Scroll to explore
/about

I'm an AI/ML engineer who loves the distance between a research paper and a product people rely on — and closing it. From data engineering and fine-tuning to deploying voice assistants and LLM agents at scale, I build AI systems that measurably improve how people work and decide.

I lead AI initiatives end-to-end — architecture, cloud deployment on AWS, Azure and IBM Cloud, and the monitoring that keeps them reliable — while mentoring engineers and keeping cross-functional teams aligned.

Based inChicago, Illinois, USA
NowConversational AI Full Stack Engineer, Aetna (CVS Health)
PreviouslyLead AI Developer, ImpacterAI · AI/ML Engineer, IBM
EducationM.S. AI & ML, DePaul University
7+years building AI systems
8roles across industry & research
23projects shipped & prototyped
15certifications earned
/experience

Where I've shipped

From robotics labs at IBM to voice AI for healthcare at Aetna — the most recent first.

NOWJan 2026 – PresentChicago, IL

Conversational AI Full Stack Engineer

Aetna, a CVS Health Company
  • Own end-to-end design of voice-based virtual assistants for healthcare provider and member services — conversational flows, intent models and dialogue logic for production voice navigators.
  • Designed reusable conversation components and dialog templates that cut development time for new healthcare use cases by 35%.
  • Improved intent recognition by analysing production conversations and expanding training data, reducing fallback responses by 22%.
  • Develop backend integrations in Python and JavaScript connecting conversational systems to enterprise APIs; lead NLP training, intent tuning and speech-to-text optimisation.
  • Integrated conversational services with enterprise systems through REST APIs for secure retrieval of member eligibility, claims and provider information.
  • The go-to engineer for production debugging — diagnosing error patterns, improving logging and reducing time-to-resolution across voice AI pipelines.
  • Implemented conversation analytics dashboards monitoring containment rate, user satisfaction and intent accuracy across production deployments.
  • Built automated testing pipelines that validate intents, entities and API responses before every production release.
  • Collaborated with UX designers on voice prompts, dialog transitions and error recovery, increasing successful task completion rates.
  • Champion clean coding practices and CI/CD discipline, partnering with frontend, backend and product stakeholders to ship reliable conversational systems.
  • Participate in architecture reviews and technical planning for enterprise conversational AI initiatives.
Jul 2025 – Jan 2026San Francisco, CA

Lead AI Developer

ImpacterAI
  • Designed and built the SARA AI autonomous sales platform with agentic AI — lead qualification, proposal generation, pricing negotiation and CRM updates with minimal human intervention.
  • Developed RAG pipelines with LangChain and Pinecone over product catalogs, pricing rules and contracts, reducing hallucinations by nearly 40%.
  • Automated deployment with Docker, Kubernetes, GitHub Actions and Azure, reducing release time by 45%.
  • Developed Python backend services and Flask APIs for agent orchestration, function calling and external system integrations.
  • Integrated Salesforce, HubSpot and internal business APIs to automate opportunity creation, meeting scheduling and customer activity tracking.
  • Designed reinforcement-learning-based multi-agent orchestration strategies to optimise agent decision-making in LLM-driven enterprise workflows.
  • Led an AI-driven predictive maintenance system for enterprise storage using TensorFlow, improving reliability and reducing downtime.
  • Identified high-value AI/ML integration points with cross-functional teams, driving a 30% improvement in data retrieval times.
  • Built a production MLOps framework with Docker and Kubernetes that streamlined deployment and monitoring within existing CI/CD.
  • Built monitoring dashboards tracking conversion rates, response latency, fallback frequency and deal progression for faster troubleshooting.
  • Led model validation and interpretability analysis to ensure fairness compliance and strengthen stakeholder trust.
  • Built an automated feature-engineering pipeline that cut manual preprocessing and accelerated the training lifecycle.
  • Mentored junior engineers on AI/ML best practice through regular knowledge-sharing sessions.
Jan 2024 – Jun 2025Chicago, IL

Graduate Research Assistant, AI/ML Engineer · Graduate TA

DePaul University
  • Developed a multi-modal (text + image) recommendation system with BERT and CLIP, boosting user engagement by 20% through Pinecone-optimised vector search.
  • Built and fine-tuned GPT models for automated content generation, cutting content-creation time by 30% while holding 95% accuracy on sentiment analysis.
  • Automated model deployment with Docker and Kubernetes, reducing inference latency by 25% for real-time computer vision applications.
  • Designed RAG evaluation pipelines comparing retrieval quality across embedding models and chunking strategies.
  • Fine-tuned open-source LLMs on domain-specific datasets to improve response quality for research applications.
  • Wrote automated benchmarks comparing latency, retrieval precision and hallucination rates across multiple LLMs.
  • Built scalable inference pipelines on AWS with SageMaker, EC2 and MLflow for reproducible experimentation.
  • Collaborated with faculty and graduate researchers to publish findings and document reproducible AI workflows.
  • Taught and led lab sections for graduate-level Python as a Graduate Teaching Assistant, mentoring students on ML fundamentals, prompt engineering and model evaluation.
Feb 2021 – Aug 2023Bangalore, India

AI/ML Engineer

IBM
  • Built generative and conversational AI with IBM WatsonX (Orchestrate, Assistant, Discovery, Speech-to-Text/Text-to-Speech), including an IT helpdesk chatbot and a hybrid document Q&A system that improved answer accuracy by 30%.
  • Integrated computer vision models into Boston Dynamics Spot robot initiatives for autonomous inspection and operational monitoring.
  • Developed CNN-based thermal hotspot detection for solar panels and thermal analysis models for Toyota manufacturing quality inspection.
  • Delivered end-to-end AI/ML solutions across conversational AI, computer vision and applied data science, from model development to enterprise deployment.
  • Contributed to Robotics, IoT and Industry 4.0 smart-manufacturing initiatives, applying AI/ML and edge computing to automate operational workflows.
  • Developed WatsonX Orchestrate skills and custom actions integrating enterprise APIs for automated workflows across business systems.
  • Ran prompt engineering and model evaluation experiments in WatsonX.ai Studio with product, backend and manufacturing stakeholders.
  • Implemented Flask backend APIs exposing AI models to enterprise applications and internal dashboards.
  • Contributed to internal AI accelerators and reusable ML components that reduced development effort across enterprise projects.
Dec 2020 – May 2021Bangalore, India

ML Project Mentor

Script Winter of Code
  • Guided developers on production-ready machine learning for open-source projects, aligning outcomes with research goals and recommendation systems.
  • Conducted code reviews to ensure scalability, managed project milestones and tracked student success.
Aug 2020 – Dec 2020Remote

Machine Learning Intern

Technocolabs
  • Converted tabular datasets into heatmaps for CNN classification and researched differential privacy for structured data.
  • Benchmarked traditional versus deep models on semi-structured data using Jupyter, Scikit-learn and SHAP.
  • Authored best-practice documentation covering preprocessing and generalisation strategies.
  • Proposed embedding-based clustering strategies and used Git, Agile methodologies and issue tracking to ensure reproducibility.
  • Presented results on model fairness and reproducibility through dashboards and storytelling.
Aug 2020 – Dec 2020Bengaluru, India

Data Scientist Intern

Technocolabs
  • Trained churn models with 91% accuracy using Scikit-learn and served predictions via Flask APIs in Docker containers.
  • Built ETL pipelines using Power BI, SQL and REST APIs; created visual reports with Matplotlib and Seaborn.
  • Developed a feature selection module using mutual information and correlation measures.
Aug 2019 – Dec 2019New Delhi, India

AI/ML Intern

MedTourEasy
  • Developed a recommendation system for treatment selection and built NLP pipelines for patient report parsing.
  • Analysed trends in medical tourism, training logistic regression and decision tree models to predict outcomes.
  • Built dashboards with Plotly and Dash and exposed predictions via Flask REST APIs.
/projects

All 23 projects

Everything I've built — with demo videos and screenshots. Click any video to play it.

SmartSign – ASL to TextComputer Vision

SmartSign – ASL to Text

  • Designed a pipeline combining MediaPipe hand tracking with an LSTM network to translate American Sign Language into text in real time.
  • Ported the model to TensorFlow Lite for efficient on‑device inference on mobile devices.
RAGflix – Movie Scene RetrievalGenAI & Agents

RAGflix – Movie Scene Retrieval

  • Built a retrieval‑augmented generation system using LangChain and Pinecone that fetches relevant movie scenes based on natural language queries.
  • Engineered embeddings for dialogue and visual descriptors to improve retrieval accuracy.
ResumeRadar – GPT Resume AssistantGenAI & Agents

ResumeRadar – GPT Resume Assistant

  • Implemented OCR to parse PDF resumes and used cosine similarity to match candidate experience with job descriptions.
  • Integrated GPT to provide tailored feedback and improvement suggestions for candidates.
DeepArt – Neural Style TransferGenAI & Agents

DeepArt – Neural Style Transfer

  • Used generative adversarial networks to apply artistic styles to user‑provided photos in real time.
  • Optimised inference using TensorRT and deployed the application via a web interface for interactive use.
Watch on YouTube ↗Data, Audio & IoT

SentimentScope – Real‑Time Sentiment Dashboard

  • Built a streaming dashboard with Plotly and VADER to visualise sentiment trends across social media platforms.
  • Employed Apache Kafka to ingest tweets and real‑time metrics to provide up‑to‑the‑minute insights for brands.
UX Research AI AgentGenAI & Agents

UX Research AI Agent

  • Developed a large language model agent that automates UX research tasks such as script generation and user feedback summarisation.
  • Connected the agent to online survey tools and analysed results using vector embeddings for clustering insights.
Watch on YouTube ↗GenAI & Agents

Embodied Search LLM

  • Built a smartglass agent that leverages GPS, CLIP and YOLOv9 to help users locate items in physical spaces.
  • Combined multimodal sensing with language models to deliver spoken guidance and AR overlays.
Watch on YouTube ↗Computer Vision

Cognitive Load Balancer

  • Designed a burnout detector for developers by analysing webcam footage and typing patterns.
  • Used physiological and behavioural signals to predict stress levels and recommend break schedules.
Watch on YouTube ↗Computer Vision

Construction Site Safety: Real‑Time Detection of PPE Using YOLO

  • Implemented data gathering, labelling, model training and dynamic real‑time deployment using a YOLO‑based detector.
  • Demonstrated significant improvements in efficiency and adherence to safety protocols on construction sites.
Watch on YouTube ↗Robotics & Industrial

Advanced Defect Identification: Capacitor Inspection Using ADLINK Camera

  • Enhanced defect identification precision through high‑resolution imaging and specialised algorithms for robust quality control.
  • Leveraged AI and image processing to detect and categorise defects quickly and accurately, optimising manufacturing efficiency and reliability.
Watch on YouTube ↗Robotics & Industrial

Enhanced Autopart Classification: COBOT Integration with Intel DepthSensing Camera

  • Integrated collaborative robot (COBOT) technology with Intel DepthSensing cameras to streamline classification of motor components.
  • Used advanced image processing and AI integration to detect and categorise parts swiftly and accurately, improving production efficiency and product reliability.
Watch on YouTube ↗Robotics & Industrial

Precision Maintenance: Thermal Inspection of Toyota Motors

  • Combined high‑resolution thermal imaging with specialised algorithms to ensure robust quality control in manufacturing.
  • Applied depth‑sensing techniques to identify and categorise parts by size and shape for precise inventory management.
Watch on YouTube ↗Robotics & Industrial

Intelligent Personnel Hazard Management System

  • Employed state‑of‑the‑art sensors and cloud analytics to detect and respond to potential hazards in real time.
  • Enhanced safety protocols and optimised resource allocation through proactive risk mitigation and real‑time insights.
Watch on YouTube ↗Robotics & Industrial

High‑Tech Solar Panel Surveillance via Drone Technology

  • Integrated thermal and visual inspections to detect anomalies and ensure optimal performance of solar arrays.
  • Provided proactive maintenance insights using real‑time data and automated analysis, maximising energy production and equipment lifespan.
Watch on YouTube ↗Robotics & Industrial

Enhanced Fire Extinguisher Inspection with Boston Dynamics Spot Robot

  • Combined thermal and visual analysis with the agility of the Boston Dynamics Spot Robot for thorough safety equipment inspection.
  • Improved maintenance efficiency and compliance by detecting potential issues early in industrial and commercial settings.
Watch on YouTube ↗Robotics & Industrial

Next‑Gen Inspection: Boston Dynamics Spot Robot & IBM Maximo Suite

  • Integrated the Boston Dynamics Spot Robot with IBM Maximo Suite analytics for circuit transformer analysis.
  • Optimised efficiency and reliability through real‑time data collection, analysis and predictive maintenance.
Watch on YouTube ↗Robotics & Industrial

Revolutionising Welding Analytics: COBOT‑Enabled Smart Manufacturing for Magna

  • Leveraged collaborative robots to enhance manufacturing efficiency and precision at Magna.
  • Integrated smart analytics to optimise welding operations, improve quality control and provide real‑time insights.
Envision for Renewable Energy, Business Intelligence and AnalyticsData, Audio & IoT

Envision for Renewable Energy, Business Intelligence and Analytics

  • Harnesses Envision's advanced analytics to optimise operations and maximise efficiency in renewable energy projects.
  • Provides real‑time insights into energy production, consumption and operational performance.
  • Utilises advanced algorithms and machine learning techniques to analyse data from solar, wind, hydro and other sustainable systems.
Towards Many to Many Communications Among Blind, Deaf & Mute UsersData, Audio & IoT

Towards Many to Many Communications Among Blind, Deaf & Mute Users

  • Enables real‑time, multi‑modal interactions by leveraging IoT devices and sensors for enhanced accessibility.
Covid‑19 Outbreak Analysis, Prediction & ForecastingData, Audio & IoT

Covid‑19 Outbreak Analysis, Prediction & Forecasting

  • Gathered and processed vast amounts of data from health organisations, government reports and public datasets.
  • Applied statistical analysis and machine learning to uncover patterns and predict the progression of Covid‑19 across regions.
  • Built predictive models to forecast future cases, hospitalisations and fatalities and developed interactive dashboards to share insights.
GenAI & Agents

AI-Native Autonomous Deal Closing System

  • End-to-end automation pipeline on Azure ML using queue triggers, blob storage and ML endpoints.
  • Hybrid neural net for objection prediction using SentenceTransformer embeddings.
GenAI & Agents

Digital Art Restoration

  • Denoising diffusion model in PyTorch restoring Dunhuang murals.
Data, Audio & IoT

Speech Emotion Recognition

  • Hybrid deep learning using Librosa, MFCC and Chroma features.
/skills

The living stack

55 tools across eight disciplines — each one shown doing what it does.

Generative AI09

  • GPT
  • Claude
  • LLaMA
  • LangChain
  • RAG
  • Agentic AI
  • Fine-tuning
  • Diffusion Models
  • RLHF

Conversational AI06

  • WatsonX Orchestrate
  • Watson Assistant
  • Watson Discovery
  • Speech-to-Text / TTS
  • Intent Modeling
  • Dialogue Systems

Languages05

  • Python
  • JavaScript
  • SQL
  • Java
  • C / C++

ML & DL08

  • PyTorch
  • TensorFlow
  • Keras
  • Hugging Face
  • Scikit-Learn
  • OpenCV
  • YOLOv8
  • XGBoost

MLOps09

  • MLflow
  • Docker
  • FastAPI
  • Flask
  • CI/CD
  • GitHub Actions
  • Airflow
  • Triton
  • Ray

Serving & Infra04

  • Kubernetes
  • ONNX
  • TensorRT
  • TF Serving

Cloud08

  • AWS SageMaker
  • AWS Bedrock
  • Lambda · S3 · EC2
  • Azure ML
  • Azure Functions · Blob
  • Google Cloud (GCP)
  • Compute Engine · GCS
  • IBM Cloud

Data & Vectors06

  • Spark
  • Databricks
  • FAISS
  • Pinecone
  • Redis
  • Weights & Biases
/education

Foundations

M.S. · AI & Machine Learning

DePaul University

Chicago, USA · 2023 – 2025
3.85/ 4.00 GPAAI specialisation
B.E. · Information Science & Engineering

Don Bosco Institute of Technology

Bengaluru, India · 2016 – 2020
4.00/ 4.00 GPASilver medalist · all 4 years
/certifications

Certified

  • NVIDIAGenerative AI Professional Certificate (DLI)
  • DL.AIGenerative AI with LLMs
  • DL.AIChatGPT Prompt Engineering
  • HFHugging Face Transformers
  • CohereLLM University
  • AzureAI Fundamentals (AI-900)
  • AzureAzure Fundamentals (AZ-900)
  • AzureIoT Developer (AZ-220)
  • IBMIBM Certified Spark – Level 1
  • GoogleCreate and Manage Cloud Resources
  • CourseraData Science Fundamentals
  • IBMIBM Cloud Certification
  • IBMIBM Certified Advocate – Cloud v1
  • IBMJava Full Stack Developer – SkillsBuild
  • KagglePython, Machine Learning, Deep Learning
/learning

Currently learning

What I'm training on next — including the newest skills the 2026 AI market is hiring for. new marks the latest.

Agentic AI & MCPtrending07

  • Model Context Protocol (MCP)
  • MCP servers
  • Agent-to-Agent (A2A)
  • Multi-agent orchestration
  • LangGraph
  • Tool calling
  • AGENTS.md / CLAUDE.md

Context Engineering & RAGtrending08

  • Context engineering
  • GraphRAG
  • Hybrid search
  • Rerankers
  • Long-context LLMs
  • Pinecone
  • FAISS
  • Weaviate

LLM Fine-Tuning06

  • LoRA
  • QLoRA
  • PEFT
  • DPO
  • GRPO
  • Distillation

Evals & Observabilitytrending06

  • Eval-driven development
  • LLM-as-judge
  • Guardrails
  • Evidently.ai
  • WhyLabs
  • Prometheus

Inference & ML Compilation07

  • vLLM
  • Triton Inference Server
  • TorchServe
  • Ray Serve
  • ONNX
  • TensorRT
  • TorchScript

AI Coding Agentstrending04

  • Claude Code
  • Cursor
  • Spec-driven development
  • Code review with agents

Edge AI & MLOps08

  • Jetson
  • Coral
  • TinyML
  • Small language models
  • MLflow
  • Kubeflow
  • Airflow
  • BentoML

Generative & Multimodal05

  • Diffusion models
  • Vision Transformers
  • Vision-language models
  • Real-time voice agents
  • OpenCV
/contact

Let's buildsomethingintelligent.

uathiya4@gmail.com