Project · 2026
ETTH
Encrypted Traffic Threat Hunter
Machine learning pipeline for encrypted traffic analysis and threat detection without payload decryption.
ABOUT
✤ Background
Hi, I'm Yukith M Joseph, a passionate student studying B.Tech — Computer Science & Engineering (Networks). My focus areas include Cybersecurity, Networking, Applied AI/ML, Secure Systems, Computer Vision, Intelligent Hardware / IoT. I enjoy building robust applications and exploring the intersection of technology and creativity.
✤ Education
B.Tech — Computer Science & Engineering (Networks)
Presidency University, Bengaluru | 2024–2028 expected
CGPA: 7.56 / 10
Computer NetworksNetwork SecurityArtificial IntelligenceDatabase Management SystemsData Structures & Algorithms
✤ Skills & Technologies
Cybersecurity & Networking
- 🌿 TCP/IP Networking
- 🌿 Network Security
- 🌿 TLS
- 🌿 JA3
- 🌿 JA3S
- 🌿 JA4 Fingerprinting
- 🌿 Encrypted Traffic Analysis
- 🌿 Access Control
- 🌿 DLP
- 🌿 Insider-Threat Monitoring
- 🌿 System Hardening
- 🌿 Breach-Response Protocols
Programming & Systems
- 🌿 Python
- 🌿 C
- 🌿 Embedded C
- 🌿 SQL
- 🌿 Git
- 🌿 JavaScript/TypeScript
- 🌿 IoT Architecture
AI/ML & Computer Vision
- 🌿 YOLOv8 / Ultralytics
- 🌿 scikit-learn
- 🌿 XGBoost
- 🌿 OpenCV
- 🌿 Whisper AI
- 🌿 Ollama
- 🌿 Google Gemini API
Web & Backend
- 🌿 FastAPI
- 🌿 Next.js
- 🌿 React
- 🌿 WebSockets
- 🌿 SQLite
- 🌿 JWT
- 🌿 bcrypt
Hardware / Tools
- 🌿 Arduino
- 🌿 ESP32
- 🌿 Unity Engine
- 🌿 Microsoft Office Suite
✤ Leadership
Social Media Head
InTech Club | June 2026 – Present
Organiser
InnovateX Tech Fest | Presidency University | 2024–Present
Event Head
Build Club | 2024–Present
Core Coordinator
One-O-One Club | 2024–Present
Event Head
Informatica Club | 2024–Present
Branches
Selected Works
Security systems, applied AI, a Roblox game and a client website — every one built end to end. Pick a branch, or keep scrolling through all of them.
- Project · 2026
AURA
Autonomous Unified Recognition Assistant
Real-time AI surveillance platform utilizing computer vision for tracking and behavioral analysis.
- Project · 2026
ShadowGuard
Enterprise AI Data Protection System
Defensive architecture concept/prototype for enterprise data protection, anomaly detection, and insider-threat monitoring.
- Project · 2025
Sugar AI
Offline Voice-Controlled Desktop Assistant
Fully offline desktop assistant running local speech processing and LLM pipelines.
- Project · 2026
Q-SHIELD
Continuous device trust with post-quantum-signed evidence
Authentication answers “who are you” once. Q-SHIELD keeps asking “can I still trust you”, using post-quantum-signed evidence.
- Project · 2026
JIVA
Privacy-preserving, real-time healthcare coordination
Ambulances, hospitals and patients on one event-driven mesh — without exposing a single hospital database.
- Roblox game · 2026
STREET HIERARCHY
Roblox grind / clicker / RPG simulator
Start as a nobody. Click your way to the top of the hierarchy. A Roblox game built end to end in Luau, from the economy curve to the VFX.
- Client website · 2026
T R Constructions
Website for a Bengaluru civil & building contractor
Production website for T R Constructions, a civil and building contractor executing structure, block work, finishes, interiors and water-tank works across Karnataka and Andhra Pradesh.
[ DEPLOYED SYSTEM ]
ETTH
Encrypted Traffic Threat Hunter
Machine learning pipeline for encrypted traffic analysis and threat detection without payload decryption.
Pythonscikit-learndpkt/ScapyPCAP Analysis
Verified Features
- ✓ leakage-controlled ML pipeline
- ✓ bidirectional flow reconstruction from raw PCAPs
- ✓ TLS ClientHello fingerprint extraction
- ✓ JA3
- ✓ JA3S
- ✓ JA4
- ✓ payloads are not decrypted
- ✓ deterministic SHA-256 behavioral hashing
- ✓ GroupShuffleSplit train/test isolation
- ✓ deterministic identifiers removed
- ✓ five experimental configurations
- ✓ 46/46 passing unit tests
- ✓ dataset/capture-environment confounding documented
- ✓ generalization claims appropriately scoped
RAW PCAP
FLOW RECONSTRUCTION
TLS FINGERPRINTS
BEHAVIORAL FEATURES
ML PIPELINE
THREAT DETECTED
STATUS: NORMAL
AURA
Autonomous Unified Recognition Assistant
Real-time AI surveillance platform utilizing computer vision for tracking and behavioral analysis.
Next.jsFastAPIYOLOv8WebSocketsSQLite
View RepositoryVerified Features
- ⚡team project
- ⚡real-time AI surveillance platform
- ⚡YOLOv8
- ⚡centroid-based multi-frame tracking
- ⚡NORMAL → OBSERVED → SUSPICIOUS behavioral state machine
- ⚡FastAPI/WebSocket backend
- ⚡JWT/bcrypt authentication
- ⚡polygon-based zone intrusion engine
- ⚡incident generation
- ⚡evidence snapshots
- ⚡Next.js live detection dashboard
[ EXPERIMENTAL PROTOTYPE ]
ShadowGuard
Enterprise AI Data Protection System
Defensive architecture concept/prototype for enterprise data protection, anomaly detection, and insider-threat monitoring.
CybersecurityAccess ControlDLP
Architecture Features
defensive architecture concept/prototype
access control
DLP
anomaly detection
policy enforcement
system hardening
breach-response workflow
shutdown protocols
insider-threat monitoring
PAYLOAD
IAM
DLP
POLICY
DECISION
[ OFFLINE SYSTEM ]
Sugar AI
Offline Voice-Controlled Desktop Assistant
Fully offline desktop assistant running local speech processing and LLM pipelines.
PythonWhisper AIOllamaMeloTTSCustomTkinter
View Source1. INPUTUSER VOICE
→
2. STT & LLMWHISPER + OLLAMA
→
3. OUTPUTMELOTTS
Core Capabilities
- fully offline desktop assistant
- on-device speech/language processing
- user data remains local
- multithreaded transcription/inference/playback pipeline
- responsive voice interaction
[ Post-quantum device trust ]
Hackathon prototype
Q-SHIELD
Continuous device trust with post-quantum-signed evidence
Authentication answers “who are you” once. Q-SHIELD keeps asking “can I still trust you”, using post-quantum-signed evidence.
Q-SHIELD keeps scoring every device from independent evidence — cryptographic identity, the physical tamper switch, sensor ranges, configuration integrity, network liveness and signed camera observations. When the evidence turns against a device it is quarantined at the gateway, every decision is recorded in a signed evidence chain, and the device only returns after a verified, operator-started recovery rebuilds its trust. Built as a working prototype for a hackathon; device telemetry is simulated and the ESP32 firmware is a skeleton.
Verified features
- explainable trust engine: six weighted evidence factors with hysteresis
- quarantine enforcement at the gateway, with operator-started verified recovery
- evidence chain: SHA-256 linked and ML-DSA-65 (FIPS 204) signed
- signed camera observations over an ML-KEM-768 (FIPS 203) session
- PQC checked against NIST ACVP test vectors
PythonFastAPISQLiteOpenCVYOLO11nML-KEM / ML-DSA
Continuous trust · try it
TrustedAll evidence holds — the device keeps its access.
Tap a factor to make its evidence fail.
Evidence chain · SHA-256 linked · ML-DSA-65 signed
- Device enrolled
Illustration of the behaviour described in the README. The real trust engine weighs all six factors with hysteresis.
[ Emergency coordination ]
Phase 6 (frozen) · decision authority in shadow mode
JIVA
Privacy-preserving, real-time healthcare coordination
Ambulances, hospitals and patients on one event-driven mesh — without exposing a single hospital database.
In an emergency, the nearest hospital is often not the right one. JIVA's rule is simple: never assume a bed is open. Public data says what a hospital can do; only a live, time-bounded acceptance from that hospital says it will — everything else stays UNKNOWN. Every fact is an event on a central bus, state engines project Patient, Hospital and Ambulance state, and four role-scoped apps update live.
Verified features
- event-driven core: state engines project every event into live state
- Hospital Acceptance Protocol with automatic re-routing on rejection or timeout
- every derived value carries confidence and provenance
- Care Feasibility Engine running in shadow mode only
- provider-agnostic routing: Valhalla, OSRM or a labelled synthetic fallback
TypeScriptExpressSocket.IOZodMapLibreAWS CDK
Hospital Acceptance Protocol · try it
Never assume a bed is open. Answer for the hospital the ambulance is asking:
- 1Ambulance → Hospital AUNKNOWN · awaiting reply
Four role-scoped apps · live over Socket.IO
ManagementAmbulanceHospitalPatient
Illustration of the protocol described in the README.
[ Roblox game · Luau ]
Playable alpha
STREET HIERARCHY
Roblox grind / clicker / RPG simulator
Start as a nobody. Click your way to the top of the hierarchy. A Roblox game built end to end in Luau, from the economy curve to the VFX.
An unofficial, fan-made grind / clicker / RPG simulator inspired by the power hierarchy of the Lookism webtoon. Every click is +1 Power: fight your way up a ladder of fighters, earn Trophies, upgrade Fists, hatch pets, roll Auras and Rebirth for permanent multipliers — until the final boss falls and the game turns into the Brainrot Dimension. Every fighter, pet and aura is an original, data-driven creation.
Verified features
- 72 scripts · ~13.5K lines of typed Luau
- server-authoritative: clicks batched every 0.2 s through token-bucket rate limits
- 8 zones, each with its own lighting mood, music and boss arena
- 34 data-driven R15 fighters with telegraphed NPC AI
- 49 pets across 8 eggs, 17 auras and 12 fist tiers
LuauRoblox StudioRojoServer-authoritative
[ Client website · live ]
T R Constructions
Website for a Bengaluru civil & building contractor
Production website for T R Constructions, a civil and building contractor executing structure, block work, finishes, interiors and water-tank works across Karnataka and Andhra Pradesh.
A production site built around the company's own work: a hero where the TRC mark assembles in 3D, a ground-to-finish sequence told through real sites, a sump/overhead-tank model, and a horizontal project gallery. The TRC mark is drawn from structure — a beam on a column, a braced frame and a channel section.
Verified features
- Next.js 16 App Router, TypeScript and Tailwind CSS v4
- GSAP motion: hero TRC assembly, ground-to-finish sequence, horizontal project gallery
- React Three Fiber scenes that load only on capable desktops, with SVG fallbacks
- typed content layer for projects, services, affiliations and testimonials
- static CSP and security headers; rate-limited enquiry endpoint with honeypot
Next.js 16TypeScriptTailwind CSS v4GSAPReact Three Fiber
〰EXPERIENCE〰
◈
NVIDIA × Presidency
Capstone Project Internship
Capstone collaboration bridging academic research with industry engineering.
◈
Elevance
Full-Stack Development Intern
Full-stack development internship building production systems.
✤ Achievements
✤ Achievements
Milestones, recognition, and growth rings from competitive events.
Participant
Odoo × NMIT Hackathon 2026
NITTE Meenakshi Institute of Technology
2026
Top 30 Finalist
Smart India Hackathon 2025
SIH
2025
National Finalist
Nirmith 2026
NITTE Meenakshi Institute of Technology
2026
Runner-Up
InnovateX 4.0 - Hardware Hackathon
Presidency University
2024
Runner-Up
InnovateX 4.0 - Hardware Expo
Presidency University
2024
CONNECT