Full Stack Software Engineer · Dublin, Ireland 🇮🇪 · MSc Data Analytics (NCI Dublin) · BTech CSE, Honours in AI/ML
| At Easebuzz (fintech, 22 months: Intern → SDE-2) | Impact |
|---|---|
Replaced sync partner-commission calc with Python asyncio concurrency |
−70% backend processing load on 150k+ daily transactions |
| Built mass merchant onboarding API (config + approvals + bank verification) | 3 days → 30 minutes, 8,000+ merchants, ~30% gross revenue growth |
| Modular Django REST APIs for partner–merchant configuration | 50,000+ partners served, 98% uptime |
| First AI support chat on Amazon Bedrock, multi-AZ deployment | Shipped company's first production LLM feature |
| CI/CD with Docker + Jenkins across 10 services | Deployment: 2 days → 15 minutes |
💼 Open to software engineering & data roles in Ireland · 🏆 Employee of the Quarter within 3 months · promoted twice in 22 months
The hard part: 80%+ of Alzheimer's-detection literature stops at binary classification. Extending an interpretable Transformer to 4-class staging across 44,000 MRI scans — and explaining which brain regions drove each prediction — is a different problem.
Triplet Attention for clinically-relevant explainability · A100 training with mixed precision + cosine annealing · benchmarked against 4 CNN baselines — including the honest finding that 2D-MRI-only Transformers need volumetric/multimodal input to beat CNNs. PyTorch Transformers Medical Imaging
The hard part: GPS noise in dense urban zones created 14% duplicate records, and there was no trip-level data at all — every operational insight had to be inferred from station-state changes alone.
Spark Structured Streaming over live GeoJSON with event-time windowing + watermarking · fault-tolerant Parquet/S3 + MongoDB sinks · surfaces demand hotspots, 180–200% station crowding, and rebalancing triggers in a Streamlit dashboard. Spark AWS S3 MongoDB Streamlit
The hard part: literature review makes you open 20 PDFs to find the 5 that matter. SUMBUD compresses that first pass — parsing messy, inconsistently-formatted papers into one scannable, summarized table.
Streamlit app with two modes: filter a papers dataset by domain, or upload PDFs directly · extracts metadata + abstracts via PyPDF2, summarizes each with DistilBART, and detects which ML methods a paper used · exports to CSV. Hugging Face Transformers Streamlit PyPDF2 NLTK
Core: Python · JavaScript · Django · React
Data & Streaming: Apache Spark · Kafka · Dagster · PostgreSQL · MongoDB · Redis
ML: PyTorch · TensorFlow · Keras · scikit-learn
Cloud & DevOps: AWS · Docker · Jenkins · Nginx · Linux