AmoghRamagiri
{ }
AI Engineer focused on shipping agentic AI and data systems: autonomous workflows, RAG pipelines, and intelligent automation.
building systems that think and act.
AI Engineer focused on shipping agentic AI and data systems: autonomous workflows, RAG pipelines, and intelligent automation. Master's in Data Science from George Washington University.
I care about the boring parts: reliable pipelines, honest evaluations, and interfaces that keep humans in the loop. I've worked across agentic AI, data systems, and product, and I tend to ship the thing.
Good AI engineering is 10% modeling and 90% making sure the system actually works in production.
where I've worked and studied.
A short trail of roles, research, and the occasional late-night pipeline fix.
Built an agentic ticket triage system (LangChain, Pinecone) clustering 100,000+ tickets into 120+ recurring patterns. Paired it with a RAG pipeline (Azure OpenAI) that auto-generates KB articles, cutting analyst triage effort 43%. Shipped end-to-end on a React/TypeScript/Tailwind + PostgreSQL stack, deployed on AWS EC2 and RDS.
Built a mortgage default prediction model (Random Forest) at 91% recall on high-risk applications. Led Azure (Data Factory, Synapse) migration and rebuilt 37+ Power BI dashboards on Microsoft Fabric.
Built SKU-level demand forecasting (XGBoost, 92% accuracy), cutting stockouts 13%. Engineered features from 2M+ customer records and ran A/B tests that lifted conversion 15%.
Built a multi-modal biometric auth system (DeepFace + MFCC, PyTorch) for offline Raspberry Pi deployment, +30% accuracy over baseline. Edge-to-cloud pipeline streamed records with sub-200ms latency.
Automated marketing reporting across 13+ clients (~75% less weekly effort). Real-time dashboards cut wasted ad spend 20%, and clustering identified 5 high-value cohorts that lifted conversion 12%.
CCAS Dean's Award. Coursework in ML, statistics, large-scale data systems, and deep learning.
AI/ML specialization. Coursework in algorithms, statistics, and computer vision, plus a capstone ML research project.
things I've built.
A mix of shipped products, ML research, and experiments. Each one taught me something I couldn't learn from a course.
A financial research agent that lazily ingests SEC 10-K/10-Q filings on demand and answers company questions with span-level citations back to the source. Hybrid retrieval (pgvector cosine + Postgres FTS) scoped per ticker, with validated [cN] citation markers and live SSE-streamed ingestion stages.
End-to-end ML pipeline and React dashboard that forecasts product demand and computes optimal stock levels. LightGBM/XGBoost predict 28-day unit sales at the product-store level, and an optimizer calculates Safety Stock, Reorder Point, and EOQ, with stockout-risk tiers surfaced in an interactive dashboard.
A browser extension plus local FastAPI backend that automates the job-application workflow: tailored LaTeX-compiled cover letters, application emails, LinkedIn/cold outreach, and a rubric-based fit score (0-10) ranking a JD against every resume variant on disk. Generation runs through the Claude API; PDFs compile locally with Tectonic.
An internal tool that turns a flood of IT support tickets into a clear picture of recurring issues. It clusters similar tickets, generates AI summaries per cluster, checks knowledge-base coverage and drafts missing KB articles, and organizes everything around versioned pipeline runs, with a grounded AI chat assistant over the data.
A machine learning web app that predicts whether a loan applicant is a Good or Bad credit risk using the German Credit dataset. Trains and compares multiple classifiers, ships the best-performing Extra Trees model through an interactive Streamlit app that returns instant predictions from applicant details.
An autonomous AI data-analyst platform: connect a database once and the agent maps the schema, relationships, and business meaning, then answers plain-English questions with SQL, charts, and dashboards. Includes a semantic AI memory layer (pgvector embeddings), a multi-LLM analyst agent, and a Presentation Mode that narrates dashboards live.
what I work with.
Languages, frameworks, and habits I lean on daily. No prescription, just whatever matches the problem.
let's build something.
Open to roles, collaborations, and interesting problems. Drop a line or grab a slot on my calendar.