Polyglot algorithmic trading simulation

A production-style trading platform demonstrating how modern exchange infrastructure is structured across performance-critical systems, quantitative analytics, and real-time visualization.

Architecture:

• C++ engine — Order book, price-time-priority matching, partial fills, market data simulation, TCP/Unix socket server
• Python analytics — VWAP, SMA/EMA, PnL, exposure, drawdown, risk metrics via NumPy/Pandas
• Node.js gateway — REST API, WebSocket streaming, rate limiting, request validation (TypeScript)
• React dashboard — Live market charts, order book, trade feed, portfolio metrics (Vite + Recharts)

Stack: C++ · Python · Node.js · TypeScript · React · Docker Compose · GitHub Actions · CMake

Highlights:

• Deterministic price-time-priority matching engine with partial fills
• End-to-end event flow: tick → match → trade → analytics → API → dashboard
• Full test pyramid: C++ unit · Pytest · Node test · React component · E2E
• Architecture Decision Records (ADR-001 through ADR-006)

Full-stack pharmacy billing, inventory & GST platform

A production-style multi-tenant SaaS for retail pharmacies handling money, stock, and tax records — where correctness matters more than delivery speed.

Architecture:

• Frontend — React 19, TypeScript, Vite, Tailwind v4, shadcn/ui
• Backend — Node 22, Express 5, Zod validation, JWT auth
• Database — PostgreSQL 15 via Prisma ORM
• Deployment — Docker Compose + nginx (same-origin by default)

Features:

• Point-of-sale billing with GST compliance
• Batch-level stock tracking with expiry alerts
• Multi-shop tenancy — isolated data per pharmacy via JWT-scoped shopId
• Three role levels: ADMIN · PHARMACIST · CASHIER
• Daily sales and GST reports
• HttpOnly refresh cookie with Origin-guard CSRF protection

Stack: React · TypeScript · Node.js · Express · Prisma · PostgreSQL · Docker · nginx

Highlights:

• 10-document docs/ set: PRD, architecture, data model, API reference, security, gap analysis
• GST acceptance fixtures in the test suite
• Versioned URL scheme with deprecation policy

SQL-like database engine built from scratch in C++

A lightweight database engine demonstrating core database internals — parsing, execution planning, and persistent storage — written from first principles without any external DB libraries.

Architecture:

• Parser — Tokenizes SQL input, handles CREATE, INSERT, SELECT, UPDATE, DELETE, builds structured command objects
• Executor — Routes parsed commands to appropriate storage operations
• Storage Engine — In-memory execution via std::map<string, vector<vector<string>>> with file-based persistence (data/*.table)

Features:

• CREATE TABLE, INSERT INTO, SELECT with WHERE filtering
• UPDATE and DELETE with WHERE clauses
• Loads data at startup from disk, persists on modification
• Modular separation: parser · executor · storage

Stack: C++ · Makefile

Highlights:

• Zero external dependencies — pure C++ standard library
• Custom tokenizer + command parser (not a full SQL grammar, but principled)
• Demonstrates low-level systems design often skipped in CRUD applications
• Clear roadmap: dynamic schemas, B-tree indexing, ACID transactions

Content-based recommendation system · 🔗 Live Demo

A machine-learning-powered movie recommender that suggests 5 similar films based on plot, genre, keywords, cast, and director — deployed live on Streamlit Cloud.

Architecture:

• Pipeline — Merge TMDB 5000 movies + credits datasets on title
• Feature engineering — Concatenate overview, genres, keywords, top-3 cast, director into a "tags" string
• Text normalization — Porter stemming via NLTK (e.g. "loving"/"loved" → "love")
• Vectorization — CountVectorizer (top 5,000 terms, English stop words removed)
• Similarity — Cosine similarity over a ~4,800 × 4,800 matrix
• Serving — Streamlit app with st.cache_resource for one-time model build

Features:

• Select from ~4,800 movies and get 5 recommendations instantly
• Live poster images pulled from TMDB API
• Runs without an API key (falls back to placeholders)
• First-load model build ~20s, subsequent recommendations are near-instant

Stack: Python · pandas · scikit-learn · NLTK · Streamlit · TMDB API

Highlights:

• Content-based filtering (no collaborative filtering) — no user data required
• Graceful degradation when TMDB_API_KEY is missing
• Documented limitations: duplicate-title merge bug, no precision@k metric
• Roadmap: TF-IDF upgrade, Docker, evaluation harness

• 🧠 Large Language Models (LLMs) Fine-tuning, prompt engineering, local inference

• 🔍 Retrieval-Augmented Generation (RAG) Vector DBs, embeddings, hybrid search

• ⚡ Distributed Systems Consensus, replication, sharding

• ☁️ Cloud-Native Backend Containers, service mesh, K8s operators

• 🤖 AI Infrastructure & MLOps Model serving, GPU orchestration, monitoring

• 📦 Microservices Architecture gRPC, event-driven design, API gateways

• 🚀 High-Performance Backend Lock-free structures, zero-copy I/O, eBPF

• 🔐 Systems Security TLS, auth protocols, threat modeling