I am an AI/ML and cloud engineer focused on making AI systems useful, inspectable, and secure. My projects span MCP agents, retrieval-augmented generation, knowledge validation, test automation, and adversarial audio ML.

At SUNY Polytechnic Institute, I work on applied AI research involving RAG, secure inference, deep learning, and adversarial robustness for cybersecurity-focused applications. Across my public projects, I connect working applications with source provenance, explicit permissions, evaluation, and reproducible evidence.

Current interests: local AI, agentic workflows on the Model Context Protocol (MCP), RAG evaluation, AI red teaming, knowledge reliability, cybersecurity automation, and MLOps.

A self-hostable AI workspace for document investigation and approved actions across connected tools. The current v2 release targets a small-team pilot.

• Built local Ollama inference, streamed responses, persistent conversations, and MCP tool execution with inspectable traces.
• Added workspace-scoped retrieval, document previews and versioning, durable ingestion jobs, and optional hybrid search with PostgreSQL/pgvector.
• Implemented exact, expiring action approvals, encrypted per-user connector credentials, JWT/RBAC, and audit records.

Architecture · v2 validation evidence

Python FastAPI Next.js MCP Ollama PostgreSQL pgvector Docker

A local application for tracing financial figures to source documents and seeing how corrections change a report.

• Built document import, source quotes, human fact review, and deterministic financial calculations using Python Decimal.
• Preserved dated report snapshots and source hashes so comparisons distinguish changed values from changed supporting evidence.
• Added cited retrieval with optional local Ollama inference, SQLite persistence, and Markdown/JSON exports.

Demo walkthrough · Release scope and validation

Python SQLite JavaScript Ollama Document Processing Source Provenance

A model-independent quality gate for knowledge before it reaches RAG systems and AI agents.

• Implemented deterministic checks for stale, unsupported, contradictory, and improperly classified claims, plus common prompt-injection indicators.
• Added SHA-256 source-drift checks, knowledge regression tests, and source-to-claim impact analysis.
• Exported SARIF findings, a Knowledge Bill of Materials (KBOM), and portable Markdown bundles using a Python standard-library core.

Architecture · Seeded benchmark and limits

Python GitHub Actions SARIF SHA-256 Knowledge Validation

A local testing platform connecting requirements, automated checks, execution evidence, and release decisions.

• Converts requirements into structured test cases and Gherkin with traceability back to the original requirement.
• Exercises a JWT/RBAC demo application through API, browser, BDD, security, and accessibility checks.
• Records test history and artifacts, analyzes failures and likely flaky tests, and applies configurable release gates. Optional Ollama summaries support the deterministic analysis.

Demo · Architecture

Python FastAPI Streamlit pytest Playwright Behave SQLite

An experimental speaker re-identification project studying whether prosody and rhythm complement conventional voiceprints under simulated voice disguise.

• Built a PyTorch BiGRU prosody encoder and compared it with ECAPA voiceprints and score fusion on held-out LibriSpeech speakers.
• Evaluated pitch, formant, and tempo transformations with rank-1 identification and equal-error-rate measurements.
• Published result tables and a formant sweep, including the limits of weak standalone prosody accuracy and simulated attacks.

Results and figures · Project guide

Python PyTorch SpeechBrain BiGRU Audio ML Adversarial Evaluation

A local-first RAG assistant for asking evidence-backed questions over cybersecurity logs and reports.

• Built PDF/TXT/LOG ingestion, per-user FAISS indexes, and answers with citations and the exact retrieved source excerpts.
• Added JWT-protected workflows, file hashes, SQLite audit records, authenticated system status, and CSV audit export.
• Supports local Ollama or hosted providers, Docker Compose, and automated checks. The MVP uses a demo account and a basic phrase-based injection guard.

Setup and demo flow

Python FastAPI Streamlit LangChain FAISS Ollama Docker

Mission Control

Build Dashboard

Pac-Man Contribution Run

Activity Radar

GenAI Engineer, SUNY Polytechnic Institute

January 2026 - Present

• Developing applied AI systems for contextual retrieval, summarization, and grounded question answering using LangChain, LlamaIndex, FastAPI, and PostgreSQL/PGVector.
• Building reproducible ML and deep-learning workflows for secure AI and cybersecurity research.

GenAI Systems Research Engineer, SUNY Polytechnic Institute

July 2025 - December 2025

• Built a secure offline LLM platform for privacy-sensitive environments using OpenWebUI, RAG, FastAPI, and local model serving.
• Evaluated prompt injection, jailbreak, and red-team attacks and contributed to secure API, network, encryption, and containerization strategies.

Data and Cloud Engineer, ConnX AI

April 2023 - July 2023

• Automated cloud observability and operational diagnostics using Python, Prometheus, Grafana, Nagios, Zabbix, and AWS CloudWatch.
• Developed real-time log analytics, anomaly detection, IP reputation, routing diagnostics, and containerized CI/CD workflows.

• Video and Audio Deepfake Datasets and Open Issues in Deepfake Technology
  • Forensic Sciences, 2024
• Design and Implementation of an AI Virtual Mouse Using Hand Gesture Recognition
  • Volume 14, Number 1, March 2024
• An Intelligent Way to Recognize Digits Using Convolutional Neural Networks
• Scarlett: Virtual Assistant and Browlett Browser
• Tachyon: Bike Rentals Made Easy

My deepfake-audio research included a survey of more than 30 studies covering datasets, detection methods, and open challenges in forensic AI.

• M.S., Network and Computer Security - SUNY Polytechnic Institute
• B.E., Computer Science - Methodist Engineering College
• CompTIA Security+ (SY0-701)
• AWS Academy Cloud Foundations
• Google Cloud Ready Facilitator
• Cisco DevNet, Cybersecurity, and Networking Essentials

I am interested in engineering roles where AI quality, security, and systems design matter together. I am especially excited by work involving enterprise RAG, AI agents, model evaluation, AI security, cybersecurity automation, and cloud-native ML platforms.

For collaboration or opportunities, reach me at saivirinchi103@gmail.com or connect on LinkedIn.