Dallas, TX · CTO @ Future Median · AI Engineer & Builder
I'm Hotragn — CTO of Future Median, a nonprofit expanding access to healthcare education, and an AI engineer who ships agents, graph-RAG systems, and products with live users. My operating principle is simple: AI output earns trust through verification — every system I ship carries its own gates, tests, and evaluation harnesses. The terrain behind this text is running a live A* route search — a nod to the vehicle-routing engines I've built.
WPT–01 / Selected builds
Agents that verify their own work, knowledge graphs that refuse to hallucinate, and products with live URLs. Every claim below is in the repo.
A living digital twin of Earth — real physics, real data, no fake numbers. Daily NASA satellite imagery, a physically-computed day/night terminator you can scrub through time, animated global wind from the latest GFS analysis, click-anywhere forecasts, and 1,200 real cities lighting up along the actual terminator. Plus a Mars twin with real orbital mechanics and a Virtual Earth time machine playing 6,000 years of city growth.
A local-first desktop AI assistant that grows its own skills: ask it to build one and it writes sandboxed code plus a test, proves it by running the test, and refines on failure. Model router prefers local Ollama, falls back to free cloud tiers. Persistent SQLite memory, replay/undo, honest telemetry — real numbers only.
Graph-RAG travel intelligence over a 57,877-node Neo4j knowledge graph (19,883 attractions, 352 cities), grounding Llama 3 via Snowflake Cortex. Triple-validation pipeline for dynamic Cypher generation — invalid queries blocked before execution, holding 74% accuracy in database query generation across six interfaces.
An always-available Agile coach built on Google's Agent Development Kit: four specialized agents collaborate on sprint planning, standups, and retros over real sprint data — data-driven coaching instead of gut feel. Ships with 30 passing tests.
A grounded study companion: upload slides or URLs and it generates syllabus-aligned quizzes, then grades free-text answers semantically. RAG over ChromaDB plus explicit prompt-injection defenses — built to not hallucinate, then tested for it.
Multi-agent supply-chain disruption detection and automated recovery on Elastic Agent Builder, wired together with the Model Context Protocol — built against a problem measured in $184B of annual global disruption cost.
"కొత్త నిర్మాత" — New Creator in Telugu, my first language. Turns one content idea into platform-optimized metadata for emerging creators: titles, descriptions, SEO tags, and hashtags tuned per platform, in seconds.
An AI assistant for Northeastern students with customizable response style, language, and length — built with OpenAI on Next.js/Firebase and tested with 500+ students on campus, reaching 95% query-resolution accuracy in testing.
WPT–01b / More on the map — click any card for the full brief
Distraction-free reading tool: DOM scraper + DistilBART real-time summaries, sub-100ms cached retrieval. Govt. of India copyright L-150423/2024.
Pantry management app, deployed and live. TypeScript.
One-click article summarization on the web. Live.
A pair of data dashboards — brewery analytics and astronomy data — built in React with live charts.
Job-role classifier (XGBoost/SVM/RF + SMOTE), 78%→85% accuracy, tracked in MLflow — with the dataset published open-access on Zenodo.
Convolutional autoencoder enhancing latent-print minutiae; training cut from 200 to 89 epochs via augmentation.
Travel discovery app in React — live deployment.
Swarm bot in C/Arduino with sensor + vision obstacle avoidance and IoT monitoring — built when robotics wasn't even in the curriculum.
Figma/Webflow/Framer case studies — Family Foundations app, Notion redesign, Rare Rabbit redesign.
WPT–01c / The extended reel — drag, scroll, or click for details
WPT–02 / Route so far
From my mother's computer-training institute in India, to research clusters at Northeastern, to shipping agents on three clouds.
WPT–02b / Education
WPT–03 / On the record
Peer-reviewed work, registered IP, open data, and the wins along the way.
Pettugani, H. et al. · IRJET, vol. 10, no. 11, 2023, pp. 761–774 — ablation study on adapting LLMs to low-resource languages.
Pettugani, H. · IJRPR, vol. 4, no. 12, 2023, pp. 1300–1304 · DOI 10.55248/gengpi.4.1223.123411
Copyright Office, Government of India — L-150423/2024 (with E. M. K. Reddy & Dr. K. K. Ravulakollu).
Pettugani, H. (2023) · published open-access on Zenodo; powers the Career Guidance System.
Knowledge Graph Conference (KGC) 2026 · poster-session winner.
WPT–04 / Toolkit
Comfort means I can open a file and start working — everything here has shipped in a real project above.
WPT–05 / Open channel
Building something with agents, knowledge graphs, or AI products — or want to talk about making AI genuinely useful inside an organization? My inbox is open.