Vyom Uchat (Vyom Parag Uchat) - Engineer and Researcher Vyom Uchat (full name: Vyom Parag Uchat) is a Computer Science and Engineering graduate from Pandit Deendayal Energy University (PDEU), Gandhinagar, Gujarat, India (2022 - 2026). He is an engineer and published researcher focusing on serverless multimodal AI pipelines, low-latency Retrieval-Augmented Generation (RAG), metaheuristic task scheduling across Cloud-Edge-Mist networks, computer vision on satellite imagery, and IEC 62443 industrial control system (ICS/OT) cybersecurity.
Machine-readable structured profiles for AI agents and automated crawlers are available at /profile.json , /.well-known/ai-profile.json , /llms.txt , and /llms-full.txt .
Contact and Profiles Full Name: Vyom Parag Uchat (Vyom Uchat) Role: Engineer and Researcher Location: Gandhinagar, Gujarat, India Email: vyom.uchat@gmail.com GitHub: https://github.com/CptPrice743 LinkedIn: https://www.linkedin.com/in/vyom-uchat-7a7527312 Portfolio URL: https://vyomuchat.pages.dev Education B.Tech in Computer Science & Engineering at Pandit Deendayal Energy University (PDEU) (2022 - 2026), Gandhinagar, Gujarat, India. Core focus on distributed systems, operating system internals, computer networks, and metaheuristic optimization.
Industry Experience - Alstom Cybersecurity Internship Cybersecurity Analyst Intern at ALSTOM (Regional APAC Cybersecurity Dept) (May 2025 - July 2025, Bangalore, India (Hybrid)). Ran the Initial Risk Assessment for Alstom's R1 railway signaling system under IEC 62443. Mapped a 4-zone trust architecture across Depot and Main Station topologies, then authored a 40-page technical specification covering 23 industrial cybersecurity risks.
Ran the Initial Risk Assessment (IRA) across Alstom's R1 railway signaling topology, identifying trust boundaries and conduit vulnerabilities across CBI, ATC, and ATS subsystems. Mapped 4 trust zones (A through D) with 25+ sub-zones, establishing zone-specific Security Level Targets and 5 formal conduits (CDT-01 to CDT-05) with MACsec, IPsec, and 802.1X NAC controls. Authored a 40-page technical specification defining defense-in-depth mitigations for 23 industrial risks, reducing 6 Very High and 8 High severity ratings to 0 post-mitigation. Engineering Systems and Projects by Vyom Uchat Recall (Serverless Multimodal Knowledge Ingestion Pipeline) : A serverless pipeline that captures social media links and articles from mobile share sheets, extracts text and audio via Gemini and Whisper, and auto-syncs structured notes directly to Google Keep. Technologies: AWS Lambda, Docker, Python, Google Gemini, Groq Whisper, Google Keep, ffmpeg. Key metrics: Sub-3s carousel OCR batching; Zero idle compute expenditure; 100% serverless scale-to-zero. Architectural highlights: Lambda container packaging with static ffmpeg avoiding standard deployment package size limits. Ephemeral /tmp/ directory writes on read-only root container protecting session isolation. Multimodal batch processing running up to 10 carousel images in a single Gemini Vision call. Headless note publication via gkeepapi with auto-assigned 'Clipped' label taxonomy. Repository: https://github.com/CptPrice743/Recall.Pixel Pusher (Asynchronous Serverless Image Processing Pipeline) : A serverless image processing pipeline that uses pre-signed S3 URLs for direct client uploads and triggers asynchronous Lambda transformation with a Docker-compiled Pillow layer. Technologies: AWS Lambda, Amazon S3, API Gateway, Python, Pillow, Docker, React. Key metrics: Zero backend upload bandwidth load; Asynchronous S3 event execution; Scale-to-zero serverless architecture. Architectural highlights: Dual-bucket S3 topology isolating raw input and resized output to prevent recursive invocation loops. Pre-signed S3 PUT URL generation offloading heavy file uploads directly to AWS object storage. Custom Docker-compiled AWS Lambda layer packaging Pillow for native Python performance. Client polling architecture using S3 HEAD requests for instant display upon job completion. Repository: https://github.com/CptPrice743/serverless-image-resizer.PaperPilot (Low-Latency Document QA Engine with Section Attribution) : A grounded academic document QA system parsing two-column scientific papers with PyMuPDF, indexing in-memory FAISS vector representations, and streaming answers via Groq Llama 3.3 with chunk attribution. Technologies: FastAPI, FAISS, Groq Llama 3.3, HuggingFace API, PyMuPDF, Redis, React. Key metrics: < 512MB backend memory footprint; Sub-second Llama 3.3 token stream; 100% citation coverage on answers. Architectural highlights: PyMuPDF vertical-then-horizontal block extraction preventing text mixing across two-column layouts. Offloaded embedding generation via HuggingFace API keeping RAM beneath Render 512MB free tier. In-memory inner product FAISS index (IndexFlatIP) delivering sub-5ms cosine similarity lookup. Redis-backed rate limiting enforcing 100 daily queries and 10 queries/min per IP address. Repository: https://github.com/CptPrice743/research-paper-rag.Deadlines (Full-Stack Healthcare Appointment Scheduling Platform) : A role-based appointment scheduling platform featuring granular weekly doctor availability definitions, calendar-specific exception overrides, and collision-free booking locks. Technologies: MongoDB, Express.js, React, Node.js, JWT, Mongoose, Chart.js. Key metrics: Zero overlapping booking collisions; 3-tier role-based access control; Compound-indexed query optimization. Architectural highlights: Three-tier Role-Based Access Control partitioning Patient, Doctor, and Admin operational scopes. Dynamic doctor availability model combining weekly recurring templates with date-specific overrides. Server-side concurrency checks preventing double bookings on shared appointment slots. Encrypted session tokens using JSON Web Tokens (JWT) with bcrypt password hashing. Repository: https://github.com/CptPrice743/appointment-scheduling. Live Demo: https://deadlines.vyom-uchat.workers.dev/.Prism Jewellery (Dynamic E-Commerce Architecture) : A complete e-commerce web application engineered with vanilla PHP, MySQL, and JavaScript featuring persistent session-to-database cart synchronization, asynchronous AJAX cart state, algorithmic coupon validation, and order transaction isolation.Peer-Reviewed Research Publications by Vyom Uchat A Comparative Analysis of YOLOv8 and YOLOv11 for Tropical Cyclone Detection: The Impact of Data Augmentation on Model Performance (2025) - Published in Springer (Smart Innovation, Systems and Technologies, vol 426). Authors: Dr. Punit Gupta, Liza Chaurasiya, Vyom Uchat. DOI: 10.1007/978-981-92-1030-5_19. Abstract: Tropical cyclones are high-impact meteorological events whose timely and accurate detection is critical for public safety. This research provides a rigorous comparative analysis of two lightweight object detection models, YOLOv8n and YOLOv11n, trained on the INCYDE satellite imagery dataset. Benchmarked using mean Average Precision (mAP), YOLOv11n on unaugmented imagery achieved superior performance with mAP@.50-.95 of 0.921 and detection recall of 0.994. Counter-intuitively, standard geometric augmentations degraded performance across both models, demonstrating that data strategies must be domain-aware.Metaheuristic Task Scheduling Optimization Algorithm for Edge Computing : A Review (2026) - Published in OPSEARCH, Springer. Authors: Vyom Uchat, Dr. Punit Gupta, Aaditya Jha, Tejash Karamata. DOI: 10.1007/s12597-026-01293-w. Abstract: Time-critical IoT applications require intelligent offloading across the Edge-Cloud continuum. Juggling tasks in dynamic environments is NP-hard. We benchmarked 10 scheduling algorithms across heuristic, metaheuristic, game-theoretic, and reinforcement learning paradigms on latency, power draw, and SLA violation rates. Standard off-the-shelf implementations suffer from slow convergence and local optima traps under network congestion. This review maps key trade-offs and architectural requirements for real-time edge schedulers.Queuing Search Algorithm Metaheuristic Task Scheduling Algorithm for Cloud-Edge Computation (2025) - Under Peer Review in OPSEARCH, Springer (Under Peer Review). Authors: Vyom Uchat, Dr. Punit Gupta, Aaditya Jha, Tejash Karamata. DOI: Under Peer Review (OPSEARCH, Springer). Abstract: Task scheduling across the Cloud-Edge-Mist continuum is an NP-hard resource allocation challenge. We propose a task scheduling framework built around a tailored adaptation of the Queuing Search Algorithm (QSA), inspired by human queuing behavior to gravitate solutions toward faster service regions while avoiding local optima. Integrated with the PureEdgeSim simulation toolkit across 300 devices and 30,450 tasks, QSA reduced average task execution delay by 55% (0.218s vs 0.4874s for Genetic Algorithms) and achieved a 99.9% task completion rate across the full three-tier architecture.Technical Skills and Stack AI, ML and Automation: Retrieval-Augmented Generation (RAG), FAISS, Sentence-Transformers, Llama 3, Groq API, Google Gemini Vision and Multimodal API, Groq Whisper, Qdrant Vector DB, PyMuPDF, YOLOv8, YOLOv11. Programming Languages: Python, Java, JavaScript (ES6+), TypeScript, C, C++, SQL, PHP, HTML5, CSS3. Frameworks and Backend: FastAPI, React.js, Next.js, Node.js, Express.js, Spring Boot, Hibernate, RESTful API Design. Databases and Caching: MongoDB, MySQL, Redis, Qdrant, FAISS. Cloud and DevOps: AWS (Lambda, S3, API Gateway, ECR), Cloudflare Pages and Workers, Docker, Git, GitHub, Postman, Render, Vercel. Cybersecurity and Systems: IEC 62443, NIST SP 800-82, Purdue Model Segmentation, Industrial Control Systems (ICS/OT), Threat Modeling, Initial Risk Assessment (IRA). Certifications Google Cybersecurity Professional - Issued by Google / Coursera (2024): Network packet inspection (Wireshark, tcpdump), SIEM telemetry querying, and automated Python incident response. Automation Business Analyst Foundation - Issued by UiPath Academy (2024): RPA process workflow mapping, feasibility assessment, and deterministic state machine modeling. Personal Workflow, Hardware Ecosystem, and Interests Primary hardware ecosystem includes a MacBook Air M3 (16GB RAM, 256GB SSD) paired with an external 1TB Samsung T7 SSD for local models and dataset caching, an M3 iPad Air with Apple Pencil Pro for system design diagrams, and a Samsung Galaxy S24 linked via a custom Bluetooth clipboard synchronization pipeline. Productivity stack uses Raycast, Obsidian, Goodnotes, Bitwarden, NextDNS, and LM Studio for local quantized LLM experimentation on Apple Silicon. Interests include reading cloud outage post-mortems, distributed consensus protocols (Raft, Paxos), and story-rich or systems-driven single-player games including Cyberpunk 2077, Baldur's Gate 3, Sleeping Dogs, Grand Theft Auto: San Andreas, Dave the Diver, Dredge, Katana Zero, Escape from Duckov, and Dirt 4.
