publications+projects: link TechRxiv preprint, drop SLAM/legacy entries, link MQTT repos

resume.tex

@@ -123,33 +123,29 @@
   \resumeSubHeadingListEnd
 
 
+%-----------PUBLICATIONS-----------------
+\section{Publications}
+  \resumeSubHeadingListStartBullets
+    \resumeSubItem{\href{https://doi.org/10.36227/techrxiv.175393256.68603080/v1}{Rule-based Tensor Mutations Embedded within LLMs for Low-Cost Mathematical Computation}}
+      {Ayyalasomayajula, S. (2025). TechRxiv preprint (IEEE). Sole-authored architecture embedding deterministic, GPU-optimized fixed-index tensor-mutation modules inside transformer LLMs (LLaMA 3B) for low-latency, step-wise-exact arithmetic and linear algebra --- improving mathematical-reasoning benchmarks without retraining or external symbolic engines. \texttt{doi:10.36227/techrxiv.175393256.68603080/v1}}
+  \resumeSubHeadingListEnd
+
+
 %-----------PROJECTS-----------------
 
 \section{Projects}
   \resumeSubHeadingListStartBullets
 
-        
-    \resumeSubItem{Preprint: Rule-based Tensor Mutations Embedded within LLMs for Low-Cost Mathematical Computation}
-      {Authored a novel architecture enhancement embedding deterministic tensor mutation modules within transformer-based LLMs (LLaMA 3B) to enable low-latency, step-wise accurate arithmetic and linear algebra computation. Leveraged GPU-optimized fixed-index tensor operations to circumvent probabilistic inference bottlenecks, achieving substantial improvements in mathematical reasoning benchmarks without retraining or external symbolic engines. Published on TechRxiv.}
   \resumeSubItem{Real-Time Quantitative Analysis Streamer}
 {Independently engineered a high-frequency financial data ingestion and analysis platform for a family quant team, sustaining \textbf{100K datapoints/s} via Kafka, Mosquitto MQTT, and DXLink streaming. Integrated Tastytrade APIs and Paho clients with a ClickHouse backend for low-latency tick storage and real-time analytics. Developed C++-based computational modules executing LSTM-driven predictive models alongside live Heston and Black–Scholes pricing for SPX, NDX, and derivative instruments. Built a Grafana observability dashboard for latency, spread, and volatility metrics, enabling continuous AI-enhanced market insight and fully automated pipeline reliability—driving a \textbf{60\% portfolio profit in Q1 2025}.}
-  \resumeSubItem{MQTT–to–Push Notification Pipeline}
+  \resumeSubItem{MQTT-to-Push Notification Pipeline \normalfont(\href{https://git.cyber.ayyalasomayajula.net/marsultor/push-notification-mqtt}{C++} $\cdot$ \href{https://git.cyber.ayyalasomayajula.net/marsultor/push-notification-mqtt-rust}{Rust} $\cdot$ \href{https://git.cyber.ayyalasomayajula.net/marsultor/news-feed-notifications}{RSS})}
-  {Developed a 3-repository system to ingest MQTT5 messages and forward them as mobile push notifications via \texttt{ntfy.sh}. Benchmarked ingestion libraries in both C++ and Rust, with Rust outperforming at \textbf{10M notifications/s} (1 KB each). Deployed a Python-based deduplicating ingest service using hash sets, which has successfully relayed 10k+ real-time news alerts to devices in the August of 2025. Designed for scalability, throughput, and reliability in high-volume messaging.}
+  {Architected a three-service notification backbone ingesting MQTT5 messages and delivering mobile push via \texttt{ntfy.sh}. Benchmarked head-to-head C++ (Paho) and Rust (rumqttc/tokio) consumers, with the async Rust rewrite sustaining \textbf{10M notifications/s} at 1KB payloads and adding tap-to-open link actions. Built the Python RSS producer with MD5-hash deduplication and QoS-1 MQTT publishing; the pipeline has relayed \textbf{10k+} real-time news alerts to devices since August 2025.}
     \resumeSubItem{Matrix FS}
       {Developed a Layer 2 distributed storage protocol atop IPFS, implemented in Rust with a performant Tauri-based Svelte frontend. Engineered end-to-end encrypted file sharing using AES-256-CBC and ChaCha20-Poly1305 AEAD schemes with deterministic tokenization for content addressing. Extended IPFS functionality with MIME type encoding to enable file-type-aware decentralized storage and retrieval. Integrated SurrealDB ticketing over Tor hidden services to facilitate privacy-preserving access control and auditability.}
 
-    \resumeSubItem{Bon Voyage (MLH Hackathon Project)}
-      {Designed and implemented an AI-powered MOOC content generator leveraging Java and Stanford CoreNLP for natural language understanding, annotation, and semantic parsing. Employed advanced NLP pipelines to extract and structure knowledge from raw educational resources, enabling dynamic course syllabus generation and personalized learning paths.}
-
-    \resumeSubItem{2022 Regional Science Fair Project}
-      {Architected an AI-driven personalized learning assistant combining OpenAI API integration and self-hosted GPT-2 models for scalable, context-aware content recommendation. Engineered multi-modal resource retrieval aggregating scientific papers, educational videos, and curated articles with semantic similarity search and relevance ranking for comprehensive knowledge dissemination.}
-
 
     \resumeSubItem{Home Infrastructure and OSS Deployment}
       {Designed, implemented, and maintain a scalable, enterprise-grade self-hosted infrastructure supporting ~50 concurrent users with partial monetization. The system architecture includes three high-availability servers orchestrating over 40 Docker containers leveraging microservices for mail servers (Postfix, Dovecot), LangChain and LangFlow NLP automation workflows, n8n event-driven automation, multicast RTP streaming clusters, WireGuard VPN mesh networking, and multiple static/dynamic web services. Contributed upstream patches and custom modules to OSS projects, while developing bespoke tooling for advanced network routing, storage management ($\ge$ 150 TB NAS with ZFS), container orchestration, and secure access gateways following zero-trust principles.}
-
-    \resumeSubItem{SLAM Guide for Autonomous Robotics}
-      {Developed an in-depth technical guide on probabilistic Simultaneous Localization and Mapping (SLAM) grounded in Durrant-Whyte’s seminal 2006 survey. Documented key algorithms such as EKF-SLAM, particle filters, and graph-based optimization methods with application insights derived from hands-on F1Tenth robotics platform experience. The guide serves as a resource bridging theory and real-time robotics system implementation.}
     \resumeSubItem{CyberPatriot Automation Scripts for USAF}
   {Developed a comprehensive suite of proprietary automation scripts in Rust and PowerShell to optimize vulnerability scanning and exploitation tasks for USAF CyberPatriot teams. The Unix-based scripts consistently contributed over 30 points per competition round by rapidly identifying and mitigating security flaws on Linux/Unix targets with sub-2-minute execution times. Windows counterparts, built in PowerShell, achieved over 25 points regularly. The toolkit was later licensed and transferred to another CyberPatriot team for ongoing development and competitive use.}
 \resumeSubItem{SSH-Driven Portfolio Site}