Harish Balaji Boominathan

Professional Experience

• Designed privacy metrics quantifying re-identification risk across structured datasets, including a normalized risk scoring system.
• Integrated image dataset support with pre-trained vision models (CLIP, ViT, DINOv2).
• Prototyped a RAG-based chatbot for interactive data readiness explanations (92% satisfaction, 1,000+ users).
• Enhanced evaluation pipeline for multi-modality and visual dashboards of quality, fairness, and privacy metrics.

• Developed multi-sensor fusion tools for 12+ nodes across Brooklyn intersections, aggregating 2+ hours of multimodal data.
• Deployed HRNet-based pose estimation on traffic video feeds (1,000+ pedestrians, safety risk analysis).
• Built ML pipelines to classify 5,000+ objects and compute pedestrian speed with 95%+ accuracy.

Featured Projects

• Multi-branch pipeline: RawAudioCNN, EfficientNetB3 (LoRA), ResNet50, PANNs-based embeddings.
• Achieved 385% throughput increase (13,100 FPS), model size 197.5MB, latency 23ms/sample (RTX6000).
• Edge deployment on Raspberry Pi 5, real-time metrics with Prometheus/Grafana (<75ms latency, 8+ users).
• Integrated MLflow for experiment tracking and edge reliability monitoring.

• Five algorithmic solvers: Greedy, Brute Force, Dynamic Programming, Constraint Programming (OR-Tools), hybrid genetic-annealing.
• Reduced planning time by 85% with smart algorithm selection and weighted preference scoring.
• Minimized API usage by 80% via caching and batch requests to Google Distance Matrix APIs.
• Route generation with Google Maps Static API, supporting multi-day, preference-weighted itineraries.

• Aggregates multilingual news from NewsAPI, deduplicates with custom NLP, ranks by relevance/recency.
• Summarizes articles using GPT-3.5 with language-aware prompts.
• Text-to-speech audio delivery via WhatsApp Business API.
• Backend: Flask, Celery, Redis for async processing, scaling, and retry logic. Monitoring via dashboards.

• XGBoost classifier (87% accuracy, 85% ROC-AUC) with SHAP interpretability.
• Streamlit dashboard for churn probability exploration (2,000+ customers).
• Feature engineering: credit score drop, transaction frequency, etc.

Education

• Research: Fine-tuned RoBERTa on 120K+ news articles with LoRA adapters (92%+ accuracy, 0.5M trainable params).
• Core: Machine Learning, Deep Learning, ML Systems Engineering, Applied Matrix Theory

• Technical Foundation: Algorithms, Operating Systems, Machine Learning, Computer Networks, Databases

Technical Expertise

Connect

This portfolio showcases the systems, models, and tools I've architected—from privacy-focused ML infrastructure to scalable backend pipelines and edge-deployed intelligence. Explore the repositories or reach out for collaboration opportunities.