Building intelligent systems is a lot like laying redstone in Minecraft: if one repeater is off or a single wire loops back on itself, the whole contraption breaks.

My focus is on creating autonomous AI agents that solve real tasks, real-time voice assistants that listen and speak as naturally as a phone call, and bulletproof security so the AI stays safely inside its sandbox.

"Understand the task. Reason what comes next. Commit the block."

Teaming up, leveling up skills, and turning fresh ideas into reliable, working software.

GUILD BASE   :: Muthayammal Engineering College (Rasipuram)
DEGREE       :: B.E. Computer Science & Engineering (Cyber Security)
WORLD CYCLE  :: Batch 2023 — 2027 (Final Year)
COORDINATES  :: Salem, Tamil Nadu, India
ACTIVE ROLE  :: Junior Software Engineer Intern @ Elverve Technologies
SPECIALTY    :: Autonomous AI Agents, Gemini Live Voice, Zero-Trust Safety
      

"Hrmmm... a solid recipe!"

Every slot in the grid has a clear purpose: Python and FastAPI run the engine, Gemini Live handles natural voice conversations, LangGraph keeps the agent on track, Neo4j and Qdrant provide memory, and Zero-Trust Security makes sure the agent only runs approved actions.

When AI agents run autonomously, they can easily get stuck in circular logic loops or hallucinate self-contradictory answers.

To prevent this, I design systems with deterministic state machines (like a clean redstone clock). The agent is given strict checkpoints: every action is verified, tool outputs are double-checked, and paradoxes are caught before they ever reach production.

Every build is crafted to solve a real challenge in the real world:

• Hospital Voice Assistant: A real-time voice AI agent for hospital appointment scheduling and patient routing. It listens and speaks simultaneously with sub-500ms delay, letting callers interrupt naturally just like talking to a real person.
  Python Gemini Live FastAPI WebRTC Exotel
[INSPECT CODEBASE]


• Image Scout (Cactus Engine): A knowledge graph engine that connects visual assets and complex questions into a searchable web of relationships using Neo4j and vector search.
  Python Neo4j Computer Vision Vector Search
[INSPECT CODEBASE]


• Portfolio Agentic Core: An interactive portfolio powered by a live AI execution loop that answers questions, queries projects, and inspects its own environment safely.
  Python LangGraph FastAPI Agentic AI
[INSPECT CODEBASE]

Just like keeping skeletons and creepers out of your base at night, an AI application must be protected from malicious prompts, unauthorized access, and unexpected inputs.

HOSTILE THREATS STRUCK DOWN:
[x] Malicious Prompt Injections & Tricky Prompts
[x] Unauthorized Tool Calls & Privilege Escalation
[x] Unsafe Code Execution & Grammar Hijacking
[x] Memory Leaks & State Boundary Breaches
DEFENSE MECHANISMS IN ACTION:
[ACTIVE] Zero-Trust Sandboxes: AI tools run with limited rights
[ACTIVE] Static Code Verification: Code is checked before running
[ACTIVE] Hardened Network Routing: Cisco-grade traffic protection

Like an Iron Golem on watch, the security boundary keeps the system stable and trustworthy 24/7.

Passing industry-recognized certifications with first-principles understanding:

[PASS] Microsoft Certified: Azure AI Solutions (AI-103)
[PASS] Microsoft Certified: Azure AI Fundamentals (AI-900)
[PASS] Cisco Certified: Network Fundamentals & Routing
      

Hardened through real labs, practical challenges, and verified exams.