Anshuman Pati
I build machine learning systems that run under real-world compute and security constraints.
Focused on edge computer vision, runtime LLM security, and deployable systems architecture. Studying Information Science and Engineering at Dayananda Sagar College of Engineering.
Technical workshop & hackathon mentorship at Dayananda Sagar College of Engineering
About
who I am and how I build
Dayananda Sagar Institutions
I focus on the intersection of machine learning research and deployable, production software.
Currently pursuing a Bachelor of Engineering in Information Science and Engineering at Dayananda Sagar College of Engineering in Bengaluru. Rather than training models in isolation, I build end-to-end pipelines that operate under strict compute, bandwidth, and security constraints.
Whether architecting multi-model edge vision networks for campus-wide anomaly detection, or analyzing activation divergence across transformer layers during runtime inference without access to weights, I care about engineering resilient systems that survive contact with real-world environments.
Python, JavaScript, C++, SQL
YOLOv8, Transformers, RAG, OpenCV, PyTorch
Next.js, Supabase, Docker, ESP32, RTSP
Experience
roles and research workAI Research Lead Intern
Technogrep Solutions LLP × DSCE (Dept. MoU)
- •Led a team of 4 to design and deploy Drishti Scene Collector, a real-time AI surveillance system running live on campus RTSP CCTV feeds, detecting medical collapses, falls, fights, crowd formation, and unauthorised vehicle entry.
- •Architected a multi-model YOLOv8 pipeline (pose, detection, segmentation variants) with mathematically-defined heuristics per anomaly class; sustained 22-25 FPS on an NVIDIA RTX 3050 (6 GB).
- •Built a Supabase cloud logging backend storing event type, anomaly score, camera ID, timestamp, and snapshot per incident, enabling post-event forensic review.
- •Authored research paper on the architecture and heuristics as primary author, presented at IEEE TEMSCON-ASPAC 2026 (forthcoming).
Freelance AI Engineer
Archscale Guild
- •Built n8n autonomous agent automations with WhatsApp API integration for streamlined client operations.
- •Engineered custom Supabase and PostgreSQL RAG pipelines for contextual domain information retrieval.
- •Implemented voice-command AI navigation capabilities with latency-optimized transcription.
- •Architected and deployed scalable backend infrastructure on Google Cloud Platform.
Tech Stack
Languages, frameworks, and edge technologies I use to build robust AI systems
YOLOv8
Computer Vision
OpenCV
Image Processing
TensorFlow / PyTorch
Deep Learning
Transformers & LLMs
NLP & Security
RAG Pipelines
Information Retrieval
Python
Core Language
JavaScript
ES6+ Modern
TypeScript
Type-Safe
C / C++
Low-Level & Systems
SQL
Relational Queries
Next.js
App Router / SSR
React
UI Library
Node.js
Runtime Environment
Tailwind CSS
Utility Styling
MongoDB Atlas
NoSQL Database
ESP32
IoT Microcontroller
Arduino Uno
Sensors & Robotics
Supabase
Cloud DB & Auth
Git & GitHub
Version Control & CI
Docker
Containerization
Featured Systems
flagship works and prototypesShatru
Runtime Backdoor Detection Engine for LLMs
A black-box runtime backdoor detection engine for large language models built in a 24-hour sprint. Shatru monitors layerwise Shannon Entropy distributions across layers 6 to 18 in Phi-3-mini to detect Trojan activation spikes without access to model weights or pre-training datasets.
Gani
IoT & Computer Vision Early-Warning Platform for Mine Safety
Predicts and detects rockfalls in open-pit mines by fusing live vibration, tilt, and seismic telemetry from ESP32 microcontroller arrays with YOLOv8 visual detection into a Next.js multi-role dashboard with severity-tiered alerts.
Drishti Scene Collector
Autonomous Real-Time AI Surveillance System for Campus Safety
Autonomous computer vision surveillance engine processing live campus RTSP feeds to detect medical collapses, falls, fights, and crowd formation. Multi-model YOLOv8 pipeline sustaining 22 to 25 FPS on a laptop RTX 3050 GPU with Supabase forensic logging.
Additional Systems & Technical Archive
Exploratory systems, hackathon builds, and embedded prototypes
Prompt efficiency optimization tool designed to reduce token consumption and carbon footprint in conversational AI models.
Real-time hand gesture recognition system built with YOLOv8 and OpenCV, fine-tuned on custom augmented gesture datasets.
Campus collaboration and event directory platform connecting students for hackathons, workshops, and inter-college contests.
Desktop credential management tool utilizing Fernet symmetric encryption derived via PBKDF2HMAC with Google Gemini integration.
Autonomous stabilization system for a physical model glider, utilizing an MPU6050 6-axis IMU for real-time pitch and roll correction.
Jellyfish-inspired underwater robot designed for marine plastic pollution detection and grasping with computer vision inference.
Automated weather protection IoT mechanism that retracts outdoor clothes-drying racks under shelter upon rain detection.
Research & Publications
peer-reviewed papers & proceedingsDrishti Scene Collector: Edge-Deployed Computer Vision Architecture for Campus Anomaly Detection
Presented at IEEE TEMSCON-ASPAC 2026
Peer-reviewed research paper presenting an edge-deployed computer vision architecture for autonomous campus safety monitoring. The work formalizes mathematically-defined heuristic models for anomalous human behavior detection (syncope/medical collapse, violent altercations, irregular crowd gathering) and evaluates multi-model YOLOv8 inference latency across live multi-camera RTSP feeds running on constrained local compute.
Talks & Leadership
speaking, mentorship and community
Hackathon Technical Defense & Evaluation
Altaria v1.0 & Genesis Club, DSCE
Defending system architecture, codebases, and edge ML pipelines before hackathon evaluation panels and industry judges. Managing judging platforms and local server infrastructure.
Technical Presentations & Dais Addresses
ACM Chapter & Department Technical Sessions
Presenting technical research on multi-model edge computer vision, runtime LLM telemetry, and edge robotics across student symposiums and technical gatherings.
Keynote EMCEE & Formal Moderation
PsychMic Club & Department Assemblies, DSCE
Stage hosting, keynote speaker introductions, and formal event moderation across institutional assemblies, flagship guest lectures, and student convocations.
Competitive Honors & Distinctions
Selected fellowships, hackathon podiums, and academic recognitions
ACM India Summer School in Edge AI & Robotics
Shatru — Runtime LLM Trojan Defense
Podium finish at Dayananda Sagar College of Engineering in a 24-hour sprint. Engineered black-box Shannon Entropy profiling across Phi-3-mini transformer layers to detect Trojan backdoor activations.
Indian Institute of Science (IISc), Bangalore · Edge intelligence, autonomous robotics, and constrained ML
Dayananda Sagar College of Engineering · Cyber & AI Track — Shatru runtime LLM backdoor detection
Ministry of Education & DSCE · IoT & Computer Vision Early-Warning Platform for Mine Safety (Gani)
MakerBlitz Hackathon · Hardware-stabilized autonomous glider with live 6-axis IMU surfaces
National Examinations · 98.2% in Grade 10 (Topper) · 95.2% in Grade 12 IP (Topper)
Honors & Awards
Recognition for academic excellence and competitive achievements
ACM India Summer School 2026
Edge AI and Robotics, IISc Bangalore
Among 100 students nationwide selected for ACM India Summer School 2026 at IISc Bangalore.
Altaria v1.0
24-hour Hackathon, DSCE
Won 1st Place in Cyber and AI track at Altaria v1.0, 24-hour hackathon.
Smart India Hackathon 2025
Government of India
Qualified for Nationals in Smart India Hackathon 2025.
MakerBlitz Hackathon
Hardware Hackathon
Sole team to integrate live electronics for hardware-stabilised glider balancing.
CBSE Grade 10
School Topper
School Topper in CBSE Grade 10 examinations.
CBSE Grade 12 (IP)
IP Topper
IP Topper in CBSE Grade 12 examinations.
Hackathons & Competitions
Projects and achievements from various hackathons and coding competitions
JAR - own everything you generate using AI
CodeUtsava 9.0 NIT Raipur
Hackman V8 Volunteer
Hackman V8
VIP Chakra - VIP threat detection system
Recursive V2
EcoAi
CypherQuest Hackathon
UniTech
ByteXync Hackathon
Roomigo - Pg Accommodation Finder
TechTrek
Internal Round Selection
Smart India Hackathon 2025
Leadership & Activities
Leadership roles and extracurricular involvement
Core Tech Team
Genesis Club (DSCE)
2025 – Present
Managed tech infrastructure and judging for Hackmanv9 (50+ teams, 200+ participants). Coordinated Coding Relay at Catalysis 4.0, the department tech fest.
EMCEE Member
PsychMic Club
2025 – Present
Member of EMCEE team for PsychMic Club events and activities.
Get in Touch
say hello or propose a projectI am open to discussions on machine learning research, software engineering roles, and applied systems engineering.
Let's Connect
Follow me on social media