Karan Manhas - Portfolio

AI, Security & Developer Tooling

• Model Context Protocol Agent: Built an MCP-based GitHub pull request summariser that parses diffs and commit history to generate structured PR summaries using LLM-assisted workflows.
  View Code on GitHub
• Natural Language to SQL Agent: Built an agent-based application using the OpenAI Agents SDK, Redshift, and Streamlit to convert natural language questions into governed SQL queries and return warehouse-backed insights.
• Vector Search R&D Environment: Set up a containerised PostgreSQL and pgvector environment for semantic search experimentation, retrieval workflows, and local AI development.

Cryptography, Benchmarking & Statistical Analysis

• CryptoMetrics: Implemented and evaluated post-quantum and classical cryptographic algorithms using C++, OpenSSL, liboqs, Python, and R. Benchmarked ML-KEM/Kyber against ECDH across execution time, memory usage, and system-level performance characteristics.
  View Code on GitHub
• CryptoGrubbs: Built a statistical analysis tool for cryptographic benchmarking data, applying Grubbs’ test to detect outliers and improve the reliability of experimental performance measurements.
  View Code on GitHub

Data Engineering & Systems

• Serverless ETL Implementations: Designed AWS Lambda-based ETL workflows supporting operational reporting across multiple users and organisations, with scheduled execution for time-sensitive business processes.
• Kimball Data Modelling: Designed dimensional models in Redshift to support enterprise analytics, reporting consistency, and downstream BI use cases.
• Flask and Redshift Dashboards: Built internal dashboards and APIs using Flask, Docker, and Redshift for monitoring operational and data warehouse resources.
• Network Telemetry Node: Built an ESP32-based telemetry node exposing health, WiFi, memory, and runtime diagnostics through lightweight HTTP endpoints and a local dashboard.
  View Code on GitHub

Applied Machine Learning

• Spam Classifier: Built a probabilistic NLP classifier using Naive Bayes and Laplace smoothing, achieving 85% classification accuracy on a held-out test set.
  View Code on GitHub
• Age and Gender Classification: Designed CNN and ResNet50-based models for simultaneous gender classification and age regression, comparing custom and transfer-learning approaches.
• Tyre Image Classification Pipeline: Developed an image-processing pipeline using OCR and feature extraction to classify tyre markings.

Applications & Tools

• ScreenTimeMonitor: Browser-based productivity tracker using HTML, JavaScript, Plotly.js, localStorage, and responsive visualisation for local screen-time analysis.
  View Live Project View Code on GitHub
• SRPN Calculator: Stack-based Reverse Polish Notation calculator implemented as a technical exercise in parsing, state management, and deterministic computation.
  View Code on GitHub

Academic Research

• MSc Dissertation: Post-Quantum Cryptography: Evaluated post-quantum and classical key exchange mechanisms in distributed environments using C++, OpenSSL, liboqs, Python, R, and AWS-based benchmarking.
• BSc Dissertation: Atmospheric Chemistry & Physics: Analysed seven years of DOAS atmospheric spectral data using R, applying time-series decomposition, regression, smoothing, and statistical modelling to pollutant concentration trends.