Prem LodhiaSoftware engineer

Architect. Build. Scale.

Software engineer building high throughput backends and low latency financial systems.

Seeking a 2027/2028 placement. Open to Summer 2027 internships.

I engineer high throughput backends, detection automation, and low latency distributed systems.

My focus is writing clean, typed, deterministic software that operates under strict latency SLAs and verifiable security guarantees.

Innovation Fest 2026
1st
1st Place, AI Elderly Care Ecosystem
First Class trajectory
88%
96% Programming, 86% Algorithms
Sentinel throughput
198tx/s
Best of 3 benchmark runs, laptop CPU
Shipped systems
12+
Distributed backends, labs and tooling

Outside engineering: native iOS development in Swift, cricket captain, and physical conditioning.

Selected work

Six systems built end to end: real time risk scoring, invoice operations, zero trust identity, network detection, an award winning care platform, and a hackathon winning local guide.

Stylised transaction stream scored by the three tier ensemble. Not live data; the benchmarks beside it are measured.

Quantitative risk and surveillance

Sentinel

Real time financial risk and anomaly detection with a three tier ensemble.

P99 latency
15.9 ms
Throughput
198 tx/s
Attack recall
98.5%

End to end P99 per benchmark run, laptop CPU

  • Run 124.16 ms
  • Run 223.08 ms
  • Run 315.88 ms
  • Target15 ms
  • Python
  • FastAPI
  • PyTorch
  • scikit-learn
  • Docker
Code Whitepaper

Financial operations engine

InvoiceOps

Local-first, AI-assisted invoice operations platform with deterministic document extraction and an auditable approval state machine.

Extraction SLA
<50 ms
Math consistency
100%
Zero-cost local-first
£0 / $0
  • Java 21 LTS
  • Spring Boot 3.4
  • Apache PDFBox
  • PostgreSQL 16
  • Docker
  • React 19
  • TypeScript
Code
The extraction pipeline above the real review workbench, where reviewers confirm every field before approval.
Stylised requests crossing each control. Policy decisions default to deny; anomalous sessions are sent to step-up auth.

Distributed backend

Zero Trust Backend

Passwordless, fail secure backend with envelope encryption and ML threat detection.

Recall (F1 0.9434)
100.0%
Benign FPR
6.00%
Crypto overhead
+2.10 ms
  • WebAuthn
  • Open Policy Agent
  • HashiCorp Vault
  • FastAPI
  • PostgreSQL
Code Whitepaper

Data pipeline

Enterprise NDR Platform

High throughput network telemetry ingestion with dual path classification and automated quarantine.

Precision
99.98%
False positives
0.04%
Real flows
100k+
  • Suricata
  • XGBoost
  • PyTorch
  • OPNsense API
  • MITRE ATT&CK
Code Whitepaper

Background: stylised flows across a network. Flagged hosts are isolated into quarantine VLAN 99, as the platform's containment step does.

Stylised simulation of the pendant at work: a quick sit-down is classified as benign, a fall crosses the impact threshold and alerts family and clinicians. The figure and signal are synthetic.

Innovation Fest 2026, 1st place

KINORA

AI powered elderly care ecosystem with fall detection, a smart pendant, a family app and a clinician dashboard.

  1. Smart pendantEdge motion telemetry and SOS
  2. Fall detection modelBenign drops vs impact falls
  3. Family appPush alerts, GPS and battery
  4. Clinician dashboardCohort trends and triage
  • IoT
  • AI
  • Hardware
  • Full stack
Stylised simulation on a 3D city map: a traveller requests a guide, nearby guides are offered the trip, the route crosses the bridge, and local events light up. The map, people and events are synthetic.

Unihack × Innovation Fest 2026, hackathon winner

Bridge

Community local guide platform connecting travellers with local guides, with an interactive map, events and recommendations.

  1. Live mapGoogle Maps with local points of interest
  2. Guide matchingTravellers paired with local guides
  3. Getting thereDirections to meet your guide
  4. Local eventsMeetups, workshops and student sessions
  • TypeScript
  • React
  • Supabase
  • Google Maps
  • Three.js

More work

01/05 Systems Diagnostics and Telemetry Toolkit

Systems Diagnostics & Telemetry Toolkit

Cross platform host diagnostic CLI that audits baselines, benchmarks performance and emits structured JSON telemetry.

  • PowerShell
  • Bash
  • Python
02/05 Hospital Management System

Hospital Management System

Clinical administration system with patient triage, doctor assignment and reports, via console and Tkinter GUI.

  • Python
  • Tkinter
  • File DB
03/05 AI Technical Interview Simulation Engine

AI Technical Interview Simulation Engine

Retrieval augmented mock interviews with multi model scoring and diagnostic reports.

  • LangChain
  • Embeddings
  • AI Agent
04/05 vulnshop-lab

Secure Web Architecture Lab

Full stack lab demonstrating the secure development lifecycle with vulnerable and remediated builds.

  • Node.js
  • Express
  • SQLite
  • AppSec
05/05 bluetrace-lab

Distributed Telemetry Ingestion Pipeline

Containerized event streaming and log aggregation with real time rule based queries.

  • Python
  • JSON-RPC
  • SIEM
  • Log Ingestion

Principles

The engineering commitments behind the work above.

Correctness and invariants

Strong domain boundaries, strict typing and defensive pre/post-conditions over loose runtime assumptions.

Predictable latency and scale

Thread-safe concurrency, memory discipline, cache locality and P99 latency SLA as first-class deliverables.

Fault-tolerant by design

Idempotent APIs under retries, distributed synchronization and structured telemetry from day one.

Work experience

Security engineering across university infrastructure, leading two student divisions, and building the front end of a 24 page civic guidance platform.

  1. Jul – Aug 2026Birmingham, UK

    • Nmap
    • Wireshark
    • Splunk
    • Active Directory
    • AWS

    Embedded within the Birmingham City University IT and security team, working on monitoring, analyzing, and safeguarding the university's enterprise infrastructure. Spearheading vulnerability assessment workflows, security monitoring, and endpoint protection schemes, while planning and executing proactive defensive measures to detect and mitigate active threats across university network boundaries.

    Security Operations & Analysis

    • Conducts regular automated and manual vulnerability scans across university servers and internal subnets to identify CVE exposures and configuration gaps
    • Monitors and analyzes security logs via centralized SIEM dashboards to detect anomalous behavior, unauthorized login attempts, and potential insider threats
    • Participates in incident response playbooks, coordinating alerts triage, isolating compromised nodes, and documenting remediation actions for internal audits
    • Performs packet capture analysis to inspect suspicious network traffic flows and identify potential command-and-control (C2) beacons or data exfiltration

    Infrastructure Hardening

    • Configures and audits firewall rulesets and ACLs to enforce least-privilege network segmentation across administrative and student subnets
    • Audits cloud access control policies and IAM roles within host environments to identify and eliminate privilege escalation risks and exposed secrets
    • Deploys security patches, updates server configurations, and hardens operating system baselines according to CIS Benchmarks
    • Enforces Endpoint Detection and Response (EDR) agent coverage and verifies policy compliance across active Windows and Linux server nodes

    Tools & Technologies

    Kali Linux, Nmap, Wireshark, Splunk, Active Directory, AWS, Firewalls & ACLs, EDR Agents, Vulnerability Scanners, CIS Benchmarks

  2. Mar – Jul 20265 months

    • AWS
    • Kali Linux
    • Burp Suite
    • Prowler

    Leading the Cyber Security Division of the BCU Student Computing Association, responsible for designing and delivering technical programmes, events, and learning experiences that give students hands-on experience in cloud security, penetration testing, and defensive security. Responsible for full curriculum design, session delivery, lab guide development, GitHub resource management, and end-to-end technical coordination across all division activities. Flagship programme: Hack Your Own Cloud — an 8-week structured cloud security programme where student teams deploy, attack, and defend real AWS environments.

    Division Leadership

    • Leads all technical programmes, workshops, and events within the Cyber Security Division from conception through to delivery
    • Designs and maintains curriculum, lab guides, and challenge resources across all division activities, hosted and managed via GitHub
    • Acts as primary technical mentor for all students in the division, supporting learning across all experience levels from complete beginners to advanced students
    • Liaises with the SCA VP Cyber Security on logistics, planning, and strategic direction of the division
    • Scouts and onboards new programme ideas, evaluating feasibility, technical requirements, and student impact before committing to delivery

    Hack Your Own Cloud — Flagship Programme

    • Designed and owns the full 8-week cloud security curriculum where students deploy deliberately vulnerable environments on AWS, attack assigned peer environments, and harden their own infrastructure
    • Built a flag-based CTF system using AWS CLI bash scripts and deterministic hash generation to verify genuine exploitation rather than theoretical knowledge
    • Architected separate AWS sub-accounts per team with budget alerts, spending caps, and VPC-level containment to ensure safe, legal, and cost-controlled peer attack exercises
    • Developed weekly lab guides covering AWS EC2, S3 misconfiguration, IAM privilege escalation, insecure APIs, network security, and cloud-native monitoring tools
    • Introduced and runs a structured peer attack challenge functioning as the division's flagship CTF event each semester

    Technical Delivery

    • Leads every weekly session using a guided learning approach — actively mentoring students through real vulnerabilities rather than delivering passive lectures
    • Implements monitoring and hardening workflows using Scout Suite, Prowler, CloudWatch, and Azure Monitor as part of the Week 7 remediation phase
    • Manages safety and containment protocols including rules of engagement, account isolation, and UK Computer Misuse Act compliance across all offensive security exercises

    Tools and Technologies

    AWS (EC2, S3, IAM, VPC, CloudWatch), Kali Linux, Nmap, Burp Suite Community, DVWA, WebGoat, Scout Suite, Prowler, AWS CLI, GitHub

  3. Mar – Jul 20265 months

    • Node.js
    • Express
    • Flask
    • React
    • Vercel

    Coordinating the Software Engineering Division of the BCU Student Computing Association, responsible for designing and delivering programmes, workshops, and events that build real technical skills in software development and engineering. Responsible for full curriculum design, session delivery, team structure, technical mentorship, and end-to-end coordination across all division activities. Flagship programme: API Workshop Series — an 8-week full stack programme where student teams build and deploy real applications powered by live external APIs.

    Division Coordination

    • Coordinates all technical programmes, workshops, and events within the Software Engineering Division from conception through to delivery
    • Designs and maintains curriculum, lab guides, and project resources across all division activities
    • Acts as primary technical mentor for all students in the division, supporting learning across beginner, intermediate, and advanced levels within the same sessions
    • Liaises with the SCA VP Software Engineering on logistics, planning, and strategic direction of the division
    • Evaluates and prioritises new programme ideas based on student impact, technical feasibility, and career relevance before committing to delivery

    API Workshop Series — Flagship Programme

    • Designed the full 8-week API Workshop Series curriculum taking students from HTTP fundamentals to a deployed full stack application using real external APIs
    • Structured a two-path backend approach (Node.js/Express or Python/Flask) to support different skill levels while maintaining consistent guided support across sessions
    • Built in a progressive difficulty curve: foundations and authentication in Weeks 1-3, backend and frontend build in Weeks 4-6, multi-API integration and deployment in Weeks 7-8
    • Designed an optional React advanced path from Week 5 onwards with additional resources provided separately so advanced students are never held back and beginners are never overwhelmed
    • Coordinates industry guest speakers and mentors to drop into build-phase sessions to review projects and share real career experience with students

    Technical Delivery

    • Leads weekly sessions using a guided learning approach — students learn by doing with active support throughout rather than passive instruction
    • Established GitHub Desktop as a core collaboration tool from Week 1, ensuring every team can manage shared codebases without conflicts from the very first session
    • Manages team formation and role assignment (frontend developer, backend developer, project lead) to give every student concrete, specific, CV-ready ownership of their contributions

    Tools and Technologies

    Node.js, Express, Python, Flask, HTML5, CSS3, JavaScript, React, Postman, GitHub Desktop, Vercel, Render, OpenWeatherMap API, GitHub API, NASA API, Spotify Web API, Open Library API

  4. Apr – Jul 2026Remote / UK

    • React
    • UI/UX
    • Admin systems
    • Database design

    Embedded as a full stack developer on CivitasAccess — a civic guidance and opportunity navigation platform built for UK students and international students. Responsible for the complete front-end architecture and UI build across 24 pages, alongside back-end planning, admin system design, and direct client collaboration throughout the product lifecycle.

    Front-End Development

    • Architected a single-file React/Babel application spanning 24 fully interactive wireframe pages including public pages, three content hubs, user account flows, and a complete admin panel
    • Built reusable component systems (navigation, card primitives, form inputs, pill labels, roadmap steps, journey nodes) maintaining consistent design tokens across the entire codebase
    • Developed an interactive Career GPS homepage flow — a multi-step guided widget allowing users to identify their current position, select a destination, and receive a personalised pathway recommendation
    • Implemented a visual journey timeline for the International Student Hub featuring clickable node-based navigation, animated progress tracking, and interactive checklists
    • Designed and built six full pathway detail pages with animated roadmap step sequences, connector lines, Career GPS sidebar panels, and linked opportunity cards
    • Built the Success Stories system — index page with category filtering, individual story pages with visual journey progression strips, and cross-linked sidebar panels

    UI/UX & Design Systems

    • Worked directly with the client to translate written briefs and conversational feedback into functional, production-ready wireframes across multiple revision cycles
    • Established and enforced a consistent brand design system: teal (#1A5C6B), gold (#93632F), typography scale, spacing grid, and component hierarchy
    • Delivered iterative wireframe audits using structured pass/gap/defer reporting to track requirement coverage and communicate scope decisions clearly
    • Proposed and agreed scope boundaries between V1 (launch) and Phase 2 (post-launch dynamic features) to protect timeline while preserving the client's long-term vision

    Back-End Planning & Admin Systems

    • Designed a full admin panel architecture covering: Opportunity CRUD management, Manage Pathways (step-by-step roadmap builder with no-code interface), Content & Stories management, User Manager, Category/Tag management, and Community Impact stats editor
    • Planned and documented the database tagging schema for opportunities, categories, and pathways to support future personalised recommendation logic
    • Structured the admin panel to allow non-technical content management — client can publish opportunities, guides, and success stories independently after handoff

    Security Considerations

    • Specified SSL certificate implementation as a deployment requirement for all data in transit
    • Designed contact form architecture to route submissions through Google Workspace email — avoiding direct server-side email exposure
    • Anticipated XSS prevention in form input handlers by verifying text sanitation requirements in the front-end
    • Documented Google SSO as the primary authentication path (supplemented by email/password) to reduce credential-attack surface in user management drafts
    • Flagged GDPR-relevant data points (e.g., student location, career tracks) for inclusion in the platform's future Privacy Policy and Terms of Service drafts
    • Separated personally identifiable information (PII) from behavioral analytics fields in the database schema draft to protect user identity
    • Proposed full network separation between admin panels and public routes to limit administrative dashboard exposure

    Client Communication

    • Authored structured client-facing update emails to summarize weekly wireframe progress, explain technical logic, and detail scope decisions
    • Created a shared "Phase 2 Deferral" document to log and agree on features that would be moved to post-launch milestones, preventing scope creep and protecting the delivery timeline

    Technical Skills Used

    • React (JSX)
    • HTML5/CSS3
    • UI/UX Design
    • Database Design
    • Security Awareness
    • Admin Systems
    • API Integration
    • Client Communication
    • Code Review

Skills

Grouped from the core outwards: languages, then the services, infrastructure and practices built on them.

Languages & core systems

  • Java (Core SE 17/21)
  • Python (FastAPI, NumPy)
  • C / C++ (RAII, Memory)
  • TypeScript
  • JavaScript
  • SQL (PostgreSQL)
  • Bash Shell
  • C#
  • PowerShell
  • Swift

Backend & distributed systems

  • Spring Boot
  • FastAPI
  • Redis (Distributed Locks)
  • Node.js & Express
  • PostgreSQL & Schemas
  • Apache PDFBox
  • Microservices Architecture
  • Idempotent API Design
  • Sliding-Window Ingestion
  • Concurrency (Atomics & Mutexes)
  • WebSockets Streaming

Cloud, systems & testing

  • Operating Systems & Linux
  • Docker Containerization
  • Kubernetes Foundations
  • AWS (EC2, S3, IAM, VPC)
  • TDD (JUnit 5, Mockito, PyTest)
  • CI/CD (GitHub Actions)
  • Cisco CCNA (IP & Routing)
  • IBM DevOps Certified
  • Git Version Control

Practices & soft skills

  • Agile & Scrum Delivery
  • Technical Mentorship (60+ Students)
  • Peer Code Review Clinics
  • Clean Architecture & Runbooks
  • Test-Driven Development (TDD)
  • Cross-Functional Collaboration
  • Problem Solving & Algorithmic Analysis
  • Continuous Learning & Growth Mindset

Applied AI & Machine Learning

  • PyTorch
  • scikit-learn
  • LangChain
  • AI-Native Tooling
  • NumPy & Pandas
  • Embedding Pipelines
  • Model Evaluation

Systems Hardening & Security

  • Google Cybersecurity Certified
  • HashiCorp Vault
  • Zero Trust Architecture
  • Open Policy Agent (OPA)
  • WebAuthn / FIDO2
  • CIS Benchmarks
  • Wireshark & Suricata

Frontend & UI Engineering

  • React
  • React Native & Expo
  • HTML5 / CSS3
  • Design Systems
  • UI/UX Primitives
  • Accessibility (WCAG AA)
  • Responsive Web

Competitions and hackathons

Building under pressure: selected competitor and award winner in engineering hackathons across the UK and Switzerland.

1st Place Winner

Innovation Fest 2026

Dean’s Award for Outstanding Engineering, Birmingham City University

Awarded 1st Place by the Dean of Computing and Engineering for designing and defending an enterprise architectural solution, evaluated on technical depth, scalability, and live defense.

  • 1st Place
  • Dean’s Award
  • System Architecture
  • Enterprise Defense

Hackathon Winner

Unihack × Innovation Fest

48-Hour Competitive Hackathon, Birmingham City University

Engineered and pitched an end-to-end full-stack software prototype under a 48-hour continuous sprint, securing 1st place against cross-university engineering teams.

  • Hackathon Winner
  • Rapid Prototyping
  • Full-Stack Sprint
  • Team Lead

International Selection

LauzHack 2026

École Polytechnique Fédérale de Lausanne (EPFL), Switzerland

Selected among competitive international applicants to attend and build at EPFL in Lausanne, Switzerland, tackling systems engineering and AI challenge tracks with European engineers.

  • Selected Competitor
  • EPFL Switzerland
  • Distributed Systems
  • AI Track

Credentials and awards

Every certificate with a public record links straight to it.

Awards
3
Certifications
15
Publicly verifiable
11
In progress
2

  • Winner

    Innovation Fest — First Place

    Birmingham City UniversityMay 2026

  • Winner

    Unihack x Innovation Fest Hackathon

    Unihack, BCUMay 2026

  • Nominee

    Most Inspirational Student of the Year

    IAMBCU AwardsApril 2026

  • Security

    Google Cybersecurity Professional Certificate

    Google, Coursera

    2026Verify
  • Security

    Ethical Hacking

    Credly

    2026Verify
  • Security

    Cisco Cybersecurity Defense Analyst

    Cisco, Credly

    2025Verify
  • Security

    Cisco Junior Cybersecurity Analyst

    Cisco, Credly

    2025Verify
  • Security

    IBM Cybersecurity Analyst

    IBM

    2025No public link
  • Cloud

    Cisco Network Technician

    Cisco, Credly

    2025Verify
  • Cloud

    AWS Certified Solutions Architect (SAA-C03)

    Amazon Web Services

    In progressIn progress
  • Cloud

    AWS Certified Cloud Support Associate

    Amazon Web Services

    Expected 2025In progress
  • AI

    Google AI Essentials

    Google, Coursera

    2026Verify
  • AI

    Fundamentals of AI Agents Using RAG & LangChain

    Vanderbilt, Coursera

    2025Verify
  • AI

    Introduction to Cloud Computing

    IBM, Coursera

    2025Verify
  • AI

    Enterprise Design Thinking Practitioner

    IBM SkillsBuild, Credly

    2025Verify
  • Engineering

    JPMorgan Chase Software Engineering Experience

    JPMorgan Chase, Forage

    2025Verify
  • Engineering

    Commonwealth Bank Cybersecurity Simulation

    Commonwealth Bank, Forage

    2025Verify
  • Engineering

    Meta iOS Developer

    Meta, Coursera

    2025No public link

Endorsements

“An exceptional learner and an asset to our classes. He demonstrated excellent analytical, practical, and critical thinking skills, allowing him to quickly understand complex computing concepts and apply them effectively.”

Dr. Hamza Mutaher Lecturer & Deputy Course Lead of Cyber Security, BCU Verify reference letter (PDF)

“An exceptionally bright student with a natural aptitude for software development and a genuine programmer's mindset. He possesses strong technical ability... I highly recommend him for any software engineering opportunity.”

Abdul Qayoom Hamal Researcher, Lecturer & Software Engineer, BCU

“Prem consistently demonstrated strong technical ability, dedication, and enthusiasm for learning. He showed excellent problem-solving skills, quickly grasped programming concepts, and applied them effectively.”

Assoc. Professor Ogerta Elezaj Associate Professor in AI & Turing Fellow, BCU

Launching soon

Cyber Interview AI

The AI interviewer built for technical and engineering disciplines.

  • 150+ roles
  • Voice or text
  • Instant report
Learn more about Cyber Interview AI

Let's build something resilient.

Open to placement for 2027/2028 and summer internships for 2027 across Software Engineering, Backend Development, Distributed Systems, Cloud Infrastructure and Systems Engineering, UK or remote.

Seeking a 2027/2028 placement. Open to Summer 2027 internships.

Case study

Real time financial risk and anomaly detection with a three tier ensemble.

P99 latency (best run)
15.9 ms
Throughput (best run)
198 tx/s
Attack recall (synthetic)
98.5%

Overview

Sentinel is a multi-model real-time risk scoring and financial surveillance engine designed for capital markets, payment processors, and investment banks. It processes real-time transaction streams, maintains stateful per-account sliding window accumulators, and scores each transaction sequentially through statistical models, tree-based isolation, and deep neural reconstruction.

Architecture & Data Pipeline

Measured Benchmark Results

End to end pipeline, three runs of the repository's benchmark on an Intel i7-1360P laptop (Windows 11, Python 3.13, CPU-only PyTorch). Results vary between runs; the 15 ms P99 target was not met.

RunP50 LatencyP95 LatencyP99 LatencyThroughput
Run 111.56 ms19.38 ms24.16 ms82.0 tx/s
Run 211.48 ms17.86 ms23.08 ms81.2 tx/s
Run 33.33 ms13.34 ms15.88 ms198.0 tx/s

Detection Quality (synthetic data)

Measured by scripts/evaluate.py on 5,000 simulated transactions (1,352 attacks): ROC-AUC 0.597, precision 0.273, recall 0.985. Recall is high, but precision is low because the simulator stamps transactions with wall-clock time, so normal traffic trips the transaction velocity rule.

  • Python
  • FastAPI
  • PyTorch
  • scikit-learn
  • Docker
  • P99 <15ms target
View GitHub repository
Case study

Modular Spring Boot 3 monolith with deterministic document extraction, 2-way PO matching, and an auditable approval workflow.

Extraction SLA
<50 ms
Math consistency
100%
Zero-cost local-first
£0 / $0

Overview & Problem Statement

Small businesses, charities, and finance teams process hundreds of supplier invoices each month across PDFs, scans, and receipts. Traditional accounts payable suffers from manual data re-entry, silent duplicate payments, undetected PO discrepancies, and unverified math errors. High-cost cloud extraction SaaS introduces vendor lock-in and monthly token fees.

InvoiceOps solves this with a 100% local-first, zero-cloud-cost platform. It pairs deterministic PDF text extraction with automated mathematical consistency checks, two-way purchase order verification, and a strictly auditable state machine ensuring human reviewers retain final financial authority.

Interactive Review Workbench

InvoiceOps Review Workbench Interface
Side-by-side document preview alongside editable extracted fields with field-level confidence ratings.

Core Architectural Invariants

SubsystemImplementationMetric / InvariantSecurity Guarantee
Modular MonolithSpring Boot 3.4.2, Java 21 LTS7 Decoupled ModulesDomain-level boundary enforcement
Document ExtractionApache PDFBox 3.0.4 + Tess4J OCR<50 ms digital PDF SLACryptographic SHA-256 duplicate rejection
Consistency EngineDeterministic Arithmetic RulesNet + Tax == Gross (0.01 tolerance)Non-blocking variance flagging
PO ReconciliationTwo-Way Commitment MatchingLine item & total PO verificationPrevents unauthorized budget overruns
Approval WorkflowEnforced State Machine4 Strict Lifecycle StatesAppend-only audit trail with actor metadata
Data & ExportPostgreSQL 16, Flyway MigrationsCWE-1236 Formula SanitizationMulti-tenant org isolation on all queries

Key Engineering Decisions

  • Assistive Automation, Not Black-Box Decisions: An invoice is never auto-approved by OCR or AI alone. Every field is annotated with confidence scores (HIGH, MEDIUM, LOW, UNVERIFIED) and human reviewers verify discrepancies.
  • Formula Injection Mitigation (CWE-1236): CSV exports sanitize spreadsheet formula triggers (=, +, -, @) to prevent remote formula execution attacks when finance teams open export sheets in Excel.
  • Strict Multi-Currency Separation: Invoices in GBP, USD, and EUR are aggregated strictly within their ISO currency codes without arbitrary currency blending.
  • Zero-Cost Local Architecture: Full operational capacity with zero external API fees, runs entirely on local Docker containers or local JVM with H2 database support.
  • Java 21 LTS
  • Spring Boot 3.4.2
  • Apache PDFBox
  • PostgreSQL 16
  • Flyway
  • React 19
  • TypeScript 6
  • Vite 8
  • Docker
  • CWE-1236 Mitigation
View GitHub repository
Case study

Passwordless, fail secure backend with envelope encryption and ML threat detection.

Recall (F1 0.9434)
100.0%
Benign FPR
6.00%
Crypto overhead
+2.10 ms

Overview

A defense-in-depth secure case-file exchange backend for security operations teams. The system implements end-to-end zero-trust controls, cryptographic data protection with HashiCorp Vault, fail-secure policy enforcement via Open Policy Agent (OPA), and behavioural identity threat detection via an IsolationForest ML pipeline. Authorization decisions default unconditionally to DENY.

Architecture & Data Flow

Fail-Secure Policy Matrix (Open Policy Agent)

ConditionTarget ActionEnforcement StateResponse
Valid Analyst Passkey + High TrustRead / Write Case FileALLOWEDHTTP 200 (Vault AES-256 Decrypted)
Anomalous Feature Vector DetectedHigh-Privilege ExportSTEP-UP REQUIREDHTTP 403 (Trigger Step-up Auth)
OPA Unreachable / TimeoutAny RequestSTRICT FAIL-SECURE DENYHTTP 503 / 403 (Default Deny)
Algorithm Confusion (alg: none)JWT Bypass AttemptBLOCKED & LOGGEDHTTP 401 (Invalid Signature)
  • WebAuthn
  • Open Policy Agent
  • HashiCorp Vault
  • FastAPI
  • PostgreSQL
  • Fail-Secure
Case study

High throughput network telemetry ingestion with dual path classification and automated quarantine.

Precision
99.98%
False positives
0.04%
Real flows ingested
100k+

Overview

Enterprise SOC network detection and containment platform. Ingests raw packet telemetry, executes dual-path classification (Suricata ET signature rules + XGBoost classifier + PyTorch benign autoencoders), and enforces automated firewall quarantine (OPNsense REST API & iptables) under a strict fail-secure guarantee.

Detection Pipeline

MITRE ATT&CK Matrix Mapping

TechniqueIDDetection EngineContainment Action
Network Service DiscoveryT1046Suricata SID 3000001BLOCK_AND_ISOLATE
Web Protocol C2 BeaconT1071.001XGBoost ClassifierQUARANTINE_VLAN 99
DNS Exfiltration TunnelT1071.004Shannon Subdomain EntropyBLOCK_AND_ISOLATE
Zero-Day Protocol TunnelT1572PyTorch Autoencoder MSEBLOCK_AND_ISOLATE
  • Suricata IDS
  • XGBoost
  • PyTorch Autoencoder
  • OPNsense API
  • MITRE ATT&CK
  • Fail-Secure
Case study

Innovation Fest 2026, 1st place

AI powered elderly care ecosystem with fall detection, a smart pendant, a family app and a clinician dashboard.

Overview

KINORA is an award-winning integrated elderly-care ecosystem designed to bridge emergency alerting, caregiver peace of mind, and clinician health telemetry. Developed for the Innovation Fest 2026 where it captured 1st Place, KINORA combines custom hardware wearable integration (smart pendant), real-time machine learning fall detection, a native family companion application, and a unified clinician dashboard.

System Components

SubsystemImplementationRole in Ecosystem
Smart Wearable PendantCustom Embedded Accelerometer / GyroscopeContinuous edge motion telemetry & SOS trigger
Fall Detection ModelEdge Inference Neural ClassifierDistinguishes benign rapid drops from actual impact falls
Family Companion AppCross-Platform Mobile ApplicationImmediate push alerting, GPS tracking & battery monitoring
Clinician Telemetry DashboardWeb Application & Analytics PortalCohort telemetry, historical trend analysis & triage reports
  • IoT
  • AI
  • Hardware Integration
  • Full stack
  • 1st Place Winner
Featured win
Philosophy

Software engineering philosophy and distributed systems architecture.

True software reliability begins by defining and enforcing system invariants from day one. In high-throughput distributed backends and low-latency engines, relying on loose runtime assumptions leads to silent state corruption, unhandled race conditions, and cascading service outages. By codifying strict schema boundaries, static and runtime invariants, and defensive validation, software stays provably correct even under degraded network states.

Modern backends handle thousands of concurrent transactions across multi-core processors. Fast code is useless if it is not thread-safe. Through disciplined memory management, cache locality, and atomic concurrency primitives (in Java Virtual Threads, Spring Boot, C++ RAII, and Redis distributed locks), systems can process high-volume workloads with deterministic memory bounds and predictable sub-10ms P99 latency.

Network boundaries and external microservices are inherently volatile. Retries, transient timeouts, and packet jitter are standard operational reality, not edge cases. Every critical endpoint must be designed idempotently, accompanied by structured metrics, distributed tracing, and clear runbooks so systems can be understood, operated, and recovered with zero data loss.