Ioan G. Istrate
Senior Backend & Platform Engineer
I build production systems
that scale.
8+ years shipping high scale web applications to tens of millions of users. Recent focus: Python concurrency, API performance, and ranking systems. Comfortable owning the full stack from data ingest to API to UX.
Experience
Tripvento
B2B hotel ranking API that scores properties by traveler intent using geospatial intelligence. 212 destinations across 25+ countries, 24K+ hotels, 200M+ pre-calculated proximity relationships. Built and scaled from first commit to 200+ destinations in under 4 months, solo.
- 4-signal ranking engine: geospatial (PostGIS ST_DWithin + GiST indexes), semantic (GPT-4o-mini for 14 traveler personas), sentiment (review analysis with dynamic weighting), and delta signals (rating velocity, price deviation). Sub-250ms API responses.
- Replaced a 55.6M-row pre-calculated edge table (11GB, 91% of DB) with on-demand spatial queries — database went from 12GB at 33 cities to 5.4GB at 212 cities, 55% smaller while scaling 6.4×.
- Multi-LLM data pipeline where models check each other’s work (City Sniffer, Hotel Auditor, Curator/Editor) — scaled from 3 cities to 212 without a QA team.
- Live MCP server for AI travel agents — LLMs and AI agents query ranking data directly for intent-based recommendations in conversational interfaces.
- Programmatic SEO generating destination guide pages; went from zero Google presence to page 1 for long-tail intent queries, scaling toward 6,000+ indexed pages.
U.S. News & World Report — Travel
Optimized Django apps serving tens of millions of users; led ranking list rollout with measurable gains.
- Improved performance on high-traffic Django applications serving tens of millions of annual users by optimizing a data management system and REST API (Django REST Framework), integrated with a React frontend and Brightspot CMS (Java/Spring Boot).
- Delivered a new ranking list and associated backend system (Django REST Framework, React frontend) that led to a significant, double-digit CTR increase to partners.
- Built multiprocessing Python scripts using Levenshtein distance mapping to address destination data gaps; identified thousands of potential high-confidence matches, far exceeding the OKR target.
- Resolved a four-year caching issue in Akamai that caused JSON lead-gen pages to malfunction; this fix repaired a primary lead generation pathway and stabilized page delivery.
- Optimized a critical Python routine from several hours down to minutes via multiprocessing and a boss/worker pattern.
- Resolved a 3rd-party API bug where a missing ‘limit’ parameter was truncating hotel inventory lists by 20%; this fix restored the full inventory and repaired a primary lead generation pathway.
Htmlthis
Operated a web development agency, managing client solutions across JavaScript, Python, PHP, and MySQL.
- Led a full stack development consultancy that specialized in building high efficiency web applications and modernizing legacy platforms.
- Led full stack migration from legacy PHP/Angular monolith to React/Node/MongoDB.
- Delivered custom solutions for diverse clients across ecommerce and marketing.
U.S. News & World Report — Travel
Optimized backend systems and frontend performance for high‑traffic travel platform.
- Significantly accelerated Tripadvisor API data retrieval by implementing a multithreading boss/worker pattern for IO-bound operations.
- Contributed to a complete responsive redesign, leading to massive, double-digit growth in yearly unique users and a significant rise in mobile traffic.
- Dramatically improved frontend performance and page load times by implementing build tool optimizations and critical CSS.
- Fixed a critical Haversine formula bug in SQL, significantly improving location-based data accuracy.
Gauge Interactive
Built e‑commerce modules for multiple client sites, integrating 3rd party APIs and handling data imports.
- Worked across 7+ client sites including moscot.com, oldtimecandy.com, dannijo.com.
- Delivered custom modules with PHP, JavaScript, MySQL for diverse e-commerce needs.
Georgia Tech (OMSCS) — CS 6200
Led ops for a 1000+ student graduate course; built Python pipeline to scrape Gradescope and generate local Git repos.
- Scrapes submission history, generates 1,000+ repos with full commit timelines.
- Detects race conditions via identical re-submissions; multiprocessing with retry logic.
Selected Projects
Concurrent C++ Job Scheduler
Thread‑safe C++17 scheduler with priority queue, atomic cancellation, graceful shutdown; ~9× faster than single‑threaded baselines.
CS 6200 ETL Grading Pipeline
Scrapes Gradescope to generate 1,000+ local Git repos with full commit history. Multiprocessing pipeline with retry logic.