Sai Praveen Tatiparthi

End-to-end deployment platform on Azure — from git push to live URL in seconds.

• Cut warm deploy times from ~6 minutes to under 12 seconds with a 5-layer caching architecture and a KEDA-autoscaled warm build-worker pool (13 ms median queue wait, 430 ms cached builds).
• Built health monitoring with version-aware auto-rollback; clients reconnect to in-flight deploys after page reload via persisted run IDs.
• Migrated production artifacts to content-addressed storage — 33 deployments, 1,085 files, zero failures.
• Drove the test suite to 1,900+ tests with enforced CI coverage gates.

Decentralized deep-RL cache coordination that beats a published ICC 2023 baseline.

• Each network router runs a Double DQN agent (N-step returns, prioritized replay) making caching decisions from local state plus a Bloom-filter neighbor signal.
• Bloom-filter coordination cuts inter-router overhead ~10x (~250 bytes/update) while raising cache hit rate to 26.65% — beating LRU+LCE (22.60%) and the ICC 2023 multi-agent DQN baseline (22.26%).
• Built the full NDN simulation stack from scratch (router CS/PIT/FIB, packet model, multi-agent environment), benchmarked from 30 to 500+ nodes with released checkpoints and reproducibility docs.

The engine running my own job search — ingests, dedupes, and LLM-scores live postings.

• Ingests live postings from Greenhouse, Lever, and Ashby APIs across 45 curated companies with content-hash + fuzzy-key deduplication into SQLite.
• Two-stage matcher: regex hard filters eliminate 64% of postings for free, then batched LLM scoring validates survivors against a structured profile with strict JSON-schema validation and per-batch retry.
• First live run: 218 postings ingested, 78 scored across 8 LLM batches with zero errors. 109 tests, TDD throughout, offline fixtures for every network/LLM boundary.

Production marketing and booking site for a salon in Livermore, CA.

• 5 routed pages, 25+ reusable components, custom scroll-reveal and parallax hooks.
• Integrated Vagaro booking and Instagram embeds; iterated over 5 months on performance, accessibility, SEO, and mobile.