AI/ML Portfolio of work by Rommel Sharma

I build applied deep learning solutions across wildlife acoustics and vision, voice AI, decision intelligence, quantitative finance, and identity systems—open, responsible, and built for real-world use.

Wildlife ID by Sound

Research-to-product pipeline for identifying Pantanal (Brazil, Bolivia, and Paraguay) biodiversity: Offline acoustic identification for 234 Pantanal species—birds, amphibians, mammals, reptiles, and insects—from short field recordings.

Tech: PyTorch, ONNX Runtime, CNN, Perch v2, Supervided and Noisy-Student training, SED pooling, test harness, MLOps

Click here to explore more →

Wildlife ID by Computer Vision

Engineered a deep-learning AI‑driven computer vision system and visual pattern framework to optimize wildlife monitoring metrics, accelerating insight processing times for multi-domain conservation projects.

Tech: PyTorch, Vision Transformers (DINOv2, MegaDescriptor‑L), GeM pooling, ArcFace, batch‑hard triplet loss, identity‑balanced mAP.

Click here to explore more →

Decision Lens

An evidence-grounded decision intelligence system that leverages Retrieval-Augmented Generation (RAG) and behavioral science literature to surface cognitive biases in high-stakes strategy.

Tech: Python, Claude for Legal Plugin, Proxy-Pointer RAG, FAISS/BM25 Hybrid Retrieval, Streamlit.

Click here to explore more →

SonarAI: Voice Synthesis

A production-grade, privacy-preserving text-to-speech engine. Features dual-path inference for consistent multilingual built-in voices and zero-shot voice cloning, outputting broadcast-ready audio.

Tech: PyTorch, XTTS v2, Kokoro-82M, FastAPI, EBU R128 Normalization, SSML

Click here to explore more →

Stocks Value Investor

A rules-driven quantitative equity screener applying Benjamin Graham and Peter Lynch review strategies to US and Indian stock markets. Features an LLM-powered investment thesis analyzer and a decoupled, sector-aware JSON rules engine.

Tech: Python, Streamlit, yfinance, Groq (LLaMA-3.3), SQLite, Docker

Click here to explore more →

Real-Time Face Recognition

A production-ready, real-time face recognition system featuring low-latency WebRTC streaming and state-of-the-art embedding extraction for secure identity verification.

Tech: InsightFace (buffalo_l), Gradio WebRTC, ONNX Runtime, HuggingFace Spaces.

Click here to explore more →

Certifications

Adobe Qualified – Adobe Experience Platform
NVIDIA-Certified Associate: AI Infrastructure and Operations
Microsoft Certified: Azure Fundamentals
Statistical Learning with Python
Oracle Certified Professional, Java SE 6 Programmer
Attention Mechanism
Google Cloud Feature Engineering
Professional Machine Learning Engineer
MLOps For Generative AI
Machine Learning Operations (MLOps) with Vertex AI: Model Evaluation
Responsible AI for Developers: Fairness & Bias
Model Armor: Securing AI Deployments