1. Interview Agent#

LangGraph · LLMs · Streamlit · Backend Systems

Designed and built a multi-phase AI interview agent that simulates a structured technical interview process rather than a single chat interaction.

Key aspects:

  • Implemented a two-phase architecture:
    • Phase 1: Resume ingestion, parsing and job description alignment
    • Phase 2: Context aware interview flow driven by agent state
  • Used LangGraph to model the interview as a state machine with explicit transitions
  • Integrated resume parsing from PDFs and dynamic job description retrieval
  • Added safety checks (guardrails) to prevent unsafe or irrelevant responses
  • Built a clean Streamlit frontend with a separate backend agent layer

This project reflects my interest in agent design, orchestration, and real-world AI system behavior not just prompt engineering.

2. Machine Learning–Based Pair Trading System#

Python · ML · Quantitative Research · Backtesting

Developed a complete algorithmic pair trading system using machine learning–driven similarity modeling instead of traditional correlation-based methods.

Core components:

  • Implemented a pseudo-Siamese neural network to learn stock pair similarity
  • Introduced Market Basket Analysis (MBA) as a pre-filtering step to reduce the search space
  • Engineered features from historical price data and trained models for pair selection
  • Built a full backtesting engine with realistic trade execution logic
  • Evaluated performance using CAGR, Sharpe ratio, win rate, and benchmark comparison

Key outcomes:

  • Achieved ~83% CAGR in long-term backtests with controlled risk
  • Demonstrated improved trade quality using MBA prefiltering
  • Outperformed benchmark returns by a significant margin

This project strengthened my understanding of ML in noisy real-world data, evaluation pitfalls and financial system constraints.

3. E-Printer — Smart Printing Optimization System#

Systems Design · Automation · Optimization

Built E-Printer, a system aimed at reducing unnecessary printing and improving resource efficiency in institutional environments.

Highlights:

  • Designed logic to optimize print requests and reduce redundant or accidental prints
  • Focused on real-world constraints such as usability, cost reduction and deployment feasibility
  • Evaluated impact based on resource savings and operational efficiency

This project was selected as a runner-up at Impetus & Contepts, validating both its technical soundness and practical relevance.

* Additional Work & Experiments#

Beyond major projects, I frequently build:

  • Backend utilities and APIs in Python
  • Small ML experiments to test ideas quickly
  • Automation scripts to reduce manual workflows

I enjoy taking ideas from rough concepts to working systems, even if they start small.

Achievements & Highlights

  • 2nd place at InC (Intercollegiate level) for the E-Printer project — a smart printing optimization system focused on reducing waste and improving resource efficiency
  • Authored and published two research papers in applied AI and system-level problem solving:
    • Pair Trading with Market Basket Filtering — published at IEEE PuneCon 2025
    • Data Deduplication using Machine Learning — published in IJSDR Journal (Vol. 10, Issue 12, December 2025)
      View Paper
  • Built multiple end-to-end projects combining AI, ML, and backend engineering
  • Strong focus on clean architecture, reliability, and real-world constraints
  • Actively preparing for full-time roles while continuing to build and learn

More projects are in progress — this terminal is a living system.