ABHISHEK VYAS
AI/ML Engineer | GenAI & Agentic Systems | Python | MLOps & Reliability
Computer Science graduate and MLOps/AI Engineer with hands-on experience in Python, Multi-Agent Systems (CrewAI), RAG pipelines, LLM reliability testing, and backend architecture. Creator of Agentic Chaos Monkey (open-source fault-injection middleware and AST vulnerability scanner for AI agents) and Jarvis (local voice-enabled AI assistant). Proven track record in containerized ML deployments, automated QA/API security testing, and production agent hardening.
J.A.R.V.I.S. // HOLOGRAPHIC COMMAND CENTER CORE
[01:44:00] JARVIS Voice Engine initialized.
[01:44:02] Connected to Ollama local endpoint (llama3.2:3b).
[01:44:03] RAG Memory loaded: 14,200 vectors in ChromaDB.
[01:44:05] Synced local Obsidian vault: d:\obsidian.
[01:44:07] Listening for user speech commands...
Competency Radar & Skill Matrix
Quantitative proficiency across AI engineering, MLOps reliability, backend languages, and security auditing.
AI & Agentic Systems
MLOps & Reliability
Languages & Web Architectures
Security & QA Auditing
Experience & Education Timeline
Professional track record in MLOps, AI application engineering, and academic background.
MLOps & AI Engineer
Nanostack Pvt Ltd, Bhopal
- Engineered Python-based AI/ML applications, local LLM inference pipelines, and Retrieval-Augmented Generation (RAG) workflows for automated text processing.
- Containerized ML services with Docker and maintained CI/CD-based automated testing and deployment pipelines.
- Developed end-to-end automated API and web-interface test suites using Python, Pytest, and Selenium WebDriver.
- Executed model API integration, endpoint stress testing, and security validation covering OWASP API risks and prompt-injection vulnerabilities.
Education Timeline
B.Tech in Computer Science
Sage University, Bhopal
Diploma in Computer Science
People's University, Bhopal
Verified Certifications
Certified Ethical Hacker (CEH)
EC-Council Certification
Machine Learning with Python
Professional Certificate
Data Analysis with Python
Data Science & Analytics
Featured Projects & Monetized Cloud SaaS
Production AI architectures, open-source reliability tooling, and cloud automation SaaS actors.
Jarvis — Local AI Assistant
Privacy-focused local AI assistant featuring an Iron Man inspired Sci-Fi React/Vite HUD interface and Node.js/Express backend for autonomous task execution. Persistent RAG-based semantic memory using vector search over developer conversation logs and local Markdown/Obsidian notes. Local Ollama LLMs with automated fallback handling, voice-command processing, and desktop telemetry.
Agentic Chaos Monkey
Open-source fault-injection and telemetry middleware simulating upstream API 502s, latency spikes, and corrupted JSON on LLM tool calls. Automated Python AST static vulnerability scanner integrated with GitHub API to detect unbounded multi-agent delegation loops (`max_iter`) and unhandled tool exceptions. Automated Resilience Scorecard quantifying retry multipliers; audited 50+ real-world repositories and caught confirmed production bug `#48` in `AgentSafeLabs/safelabs-eval` (416+ ⭐).
Bitcoin Trader AI
Real-time full-stack algorithmic trading platform with a reactive Vue.js analytics dashboard. Integrated Binance WebSockets for low-latency market streaming alongside a multi-indicator decision engine and Reinforcement Learning feedback loop.
Cloud Data & Tech Stack SaaS Actors
Production cloud automation SaaS Actors deployed on Apify Store:
• Startup Jobs & Hiring Signals Scraper (Greenhouse & Lever ATS)
• B2B Lead & Website Tech Stack Enricher
Built with high-speed async Python + HTTPX extracting real-time job postings, tech stack footprints, and contact channels for Clay and n8n pipelines.
Live Achievements & Proof-of-Work Feed
Auto-synced engineering milestones, confirmed bug catches, and live cloud releases.
Merged PR #200 in academic-commercialization-agent (771⭐): Thread Ownership & Shared Runtime Isolation
Resolved Issue #197 by overhauling cell-owned thread boundary tracking in opt-in ablation deadline mode. Isolated explicit worker and callback threads from persistent CrewAI EventBus loops, preventing false timeouts while preserving batch-halting guarantees.
Merged PR #192 in academic-commercialization-agent (771⭐): Bounded Execution Deadlines & Exit Boundary
Engineered wall-clock execution deadline harness to guard against runaway paid LLM evaluations. Built 2-stage thread cleanup (cooperative cancellation + bytecode interrupt), Pydantic Crew compatibility, honest time-incomplete usage auditing, and offline regressions.
Opened Architecture PRs on academic-commercialization-agent (771⭐) & agents-shipgate (#967)
Collaborated with maintainers on shuxiachai/academic-commercialization-agent (#191) to implement wall-clock execution deadlines, and ThreeMoonsLab/agents-shipgate (#917 / PR #967) to ship CrewAI fault-isolation patterns & offline test suites.
Root Portfolio Website & ATS Resume Release
Built and deployed standalone interactive AI Systems Architect portfolio & ATS printable resume.
Zero-False-Positive AST Engine Release
Successfully shipped zero-false-positive AST scanning rules tracking httpx.AsyncClient timeouts and test path filters.
Confirmed Vulnerability Catch in AgentSafeLabs/safelabs-eval (416+ ⭐)
Identified an unhandled JSON parsing fault trap (validate.py:308) via static AST analysis. Confirmed and patched by core maintainer in Issue #48.
Startup Jobs & Hiring Signals Scraper (Greenhouse & Lever ATS)
High-speed async Python + HTTPX cloud scraper extracting real-time startup job postings, tech stack signals, and remote indicators directly from ATS REST APIs.