Back to portfolio
Full archive
All Projects
Every project I've built — embedded systems, FPGA, AI/ML, cloud, and full-stack software — with problem, solution, contribution, and lessons learned for each. 11 projects total.
Adaptive Systolic Architecture
Problem
Edge AI accelerators need to be fast without burning a power budget FPGAs can't spare.
Solution
Designed an adaptive systolic array with dual-radix arithmetic to cut power draw while keeping ML inference throughput on target.
My Contribution
Led Verilog architecture design and Vivado verification, working alongside Gopika PL, Supriya M, and Shalini T.
Lessons Learned
Hardware-level power optimization requires trade-offs that don't show up until synthesis — simulation alone isn't enough.
MediGuard — Smart Medicine Temperature Monitor
Problem
Temperature-sensitive medicines like vaccines and insulin spoil silently in rural healthcare settings with unreliable cold storage.
Solution
Built an IoT system with DHT11/LM35 sensors, automated cooling via relay, and GSM + buzzer alerts when thresholds are breached.
My Contribution
Built the sensor and alerting pipeline in a 36-hour national hackathon with Monica Gandhi A and Swathi Govindarajan.
Lessons Learned
Designing for rural deployment means planning for intermittent connectivity from day one, not as an afterthought.
TN Election Guide
Problem
Many first-time voters in Tamil Nadu don't understand the mechanics of EVMs, VVPAT, or how government is formed after an election.
Solution
An AI-powered civic education platform with interactive learning modules, a chatbot assistant, myth-busting FAQs, and offline fallback support.
My Contribution
Coordinated feature scope and built the React front end and Gemini-powered chatbot integration under hackathon time pressure.
Lessons Learned
Scoping ruthlessly for a hackathon deadline is itself a project-management skill — cutting features early protected the core experience.
AWS Deployment — Portfolio on S3
Problem
Understanding how static hosting and cloud infrastructure actually work in production requires more than reading documentation — it needs a real, deployed project.
Solution
Deployed this portfolio as a live static website using Amazon S3 Static Website Hosting, configuring bucket policies, public access settings, and static website endpoints end-to-end.
My Contribution
Independently configured the S3 bucket, set up static website hosting, managed permissions and access policies, and deployed and validated the live site in the ap-south-1 (Mumbai) region.
Lessons Learned
Cloud deployment is as much about permissions and access policies as it is about the code itself — getting the bucket policy right mattered more than the actual file upload.
IMDb Sentiment & Emotion Analysis
Problem
Manually reading through thousands of movie reviews to gauge audience sentiment doesn't scale.
Solution
An NLP classifier combining BERT and TF-IDF, wrapped in a Streamlit UI with REST API integration, trained further using IBM Watson.
My Contribution
Built the model pipeline and UI with Reshma Sherin during an IBM internship placement.
Lessons Learned
Combining a classical approach (TF-IDF) with a transformer model gave more robust results than either alone.
IoT Patient Monitoring System
Problem
Ward staff need continuous visibility into patient vitals without manual, periodic checks.
Solution
Real-time healthcare parameter tracking built on ESP32, streaming vital signs to a live dashboard.
My Contribution
Designed the sensor-to-dashboard data pipeline with Puja S and Pavithra S.
Lessons Learned
Sensor noise handling mattered more for reliability than the dashboard UI itself.
Traffic Controller (Verilog)
Problem
Fixed-timing traffic signals don't adapt to changing intersection conditions.
Solution
An intelligent traffic control system implemented as a finite state machine in Verilog HDL.
My Contribution
Designed the FSM and simulated timing behavior with Monica Gandhi A.
Lessons Learned
State machine design forces you to enumerate every edge case up front — there's no patching it later in hardware.
Restaurant Review Sentiment Analysis
Problem
Restaurants need a fast read on customer sentiment from raw review text.
Solution
An end-to-end NLP pipeline with feature engineering and multiple ML models for classification.
My Contribution
Built the preprocessing and feature-engineering pipeline independently during the IBM internship.
Lessons Learned
Feature engineering had more impact on accuracy than swapping models.
Astable Multivibrator — PCB Design
Problem
Timer and oscillator circuits need a reliable, manufacturable square-wave source.
Solution
Designed and simulated an astable multivibrator in EasyEDA, laid out as a 2-layer PCB with through-hole components.
My Contribution
Independent circuit design, simulation, and PCB layout, prepared for JLCPCB fabrication.
Lessons Learned
Component tolerances matter far more on a real board than in simulation.
Create Your Own FM Radio Signal
Problem
Understanding FM modulation conceptually is harder without a hands-on simulation.
Solution
Simulated an FM radio station in Scilab — carrier wave modulation, audio resampling, FFT spectrum analysis, and playback validation.
My Contribution
Independent build covering the full modulation-to-playback signal chain.
Lessons Learned
FFT analysis made modulation artifacts visible that pure time-domain plots hid.
Audio Tone Generator
Problem
Generating musical tones programmatically requires precise control over waveform synthesis.
Solution
Sine wave synthesis in Scilab supporting single notes, octave scaling, chord combinations, and melody sequencing.
My Contribution
Independent build of the synthesis and sequencing logic.
Lessons Learned
Chord synthesis exposed how phase alignment affects perceived tone quality.
Temperature & Humidity Control
Problem
Environments like server rooms or greenhouses need continuous, remote climate visibility.
Solution
Real-time temperature and humidity monitoring using an ESP32 Wi-Fi module and DHT11 sensor, integrated with the Blynk IoT app.
My Contribution
Independent build of the sensor circuit and Blynk dashboard integration.
Lessons Learned
Cheap sensors need calibration routines — raw DHT11 readings drifted more than expected.
No projects match your search.