Projects — Tarique Rizwan

Portfolio

Selected
work.

AI-driven products, security layers, and experimental systems built to solve real problems.

01 · Featured

Enterprise AI security layer that detects, sanitizes, and protects sensitive information before it reaches external AI systems.

Problem: Enterprises fear leaking sensitive PII/PHI data to public LLMs, hindering AI adoption.
Solution: A proxy layer that automatically identifies and masks sensitive data using high-performance regex and LLM-based detection.
Architecture: Edge-based sanitization pipeline with real-time logging and audit trails.
Challenges: Maintaining sub-100ms latency while performing deep content inspection.
Lessons: Security at scale requires a balance between strictness and developer experience.

02 · Featured

Conversational AI financial advisor providing personalized guidance and insights.

Problem: Personal finance is complex; existing tools are either too simple or too overwhelming.
Solution: A conversational interface that simplifies financial data and delivers actionable growth insights.
Architecture: Vector database with RAG to provide context-aware financial advice.
Challenges: Ensuring numerical accuracy in financial calculations and advice.
Lessons: Clear data visualization is key to making financial guidance accessible.

03 · Featured

A modern commerce experience for pure, honest products — built around clean storytelling, fast browsing, and a frictionless checkout flow.

Problem: Small purity-first brands get lost in cluttered marketplaces where product story and trust signals disappear.
Solution: A focused storefront that leads with the product narrative, then moves visitors into a short, distraction-free purchase path.
Architecture: Component-driven React front end with typed product data, responsive layouts, and image-optimised delivery on Vercel.
Challenges: Balancing an editorial, image-heavy presentation against fast load times on mobile networks.
Lessons: Trust in commerce is designed, not declared — pacing, typography, and clarity do more than badges.

04 · Featured

A security-review workspace for phishing triage and pre-send data-leak scanning, built as a decision aid rather than an enforcement gate.

Problem: People paste sensitive text into AI tools and act on convincing phishing emails without a fast way to sanity-check either.
Solution: Two review surfaces — phishing email analysis and pre-send leak scanning — that score risk, flag indicators, and recommend next steps.
Architecture: Authenticated full-stack service with deterministic pattern detection, server-side AI enrichment on redacted content, and an incident queue for high-severity findings.
Challenges: Keeping original submissions out of storage while still producing useful, reviewable findings and alerts.
Lessons: Redaction boundaries have to be designed into the data flow from the start, not bolted on before persistence.

05 · Featured

A C++17 deep packet inspection engine that classifies and filters traffic from PCAP captures — including encrypted HTTPS flows.

Problem: Encrypted traffic hides application identity, making it hard to understand or control what a network is actually carrying.
Solution: Parses packets layer by layer and extracts the plaintext SNI field from TLS handshakes to classify 20+ applications, with rule-based blocking by IP, app, or domain.
Architecture: Raw PCAP parsing with no libpcap dependency, five-tuple flow tracking, and single- and multi-threaded builds sharing one core pipeline.
Challenges: Handling malformed and partial packets safely while keeping throughput high across worker threads.
Lessons: Working close to the bytes teaches more about protocols than any abstraction layer above them.

06 · Featured

An Android digital wellbeing app that turns raw screen-time data into understandable habits and gentle, intentional limits.

Problem: Screen-time tools answer how long you used your phone, but not which patterns are actually costing you.
Solution: Usage insight around distracting patterns and short-form video, personal boundaries, and interventions when limits are reached.
Architecture: Kotlin and Jetpack Compose with Material 3, usage-stats collection, and Gemini-powered wellbeing insights.
Challenges: Making interventions feel supportive rather than punitive while still being effective.
Lessons: Behaviour change design lives in tone and timing as much as in the underlying data.