I am a final-year Computer Science and Engineering undergraduate at VIT-AP University with a focused interest in building production-grade AI systems, agentic frameworks, and developer tooling. My engineering practice spans the full spectrum — from designing multi-LLM orchestration pipelines and agentic execution loops, to architecting cloud-native backends and contributing to open-source CLI infrastructure.

My capstone project, Rexode, is an agentic multi-LLM quality optimization and execution framework designed for AI-assisted software development. It implements a six-layer Multi-Agent Quality Enhancement (MAQE) pipeline, LangChain-based LLM routing, zero-token intent classification, and a ReAct-based agentic loop — benchmarked across GPT-4o-mini, Gemini 2.0 Flash, and Llama 3.1 8B.

I build at the intersection of AI engineering and developer experience, with a strong preference for systems that are measurable, reproducible, and defensible in production.

Open To → AI/ML Engineering Roles · Agentic Systems · LLMOps · Full Stack Engineering · Research Engineering · Open Source Collaboration

⚡ Rexode — Agentic Multi-LLM Quality Optimization Framework

An enterprise-grade agentic execution framework for AI-assisted software development, built as a fork and architectural rebrand of OpenCode CLI. Rexode introduces a six-layer Multi-Agent Quality Enhancement (MAQE) pipeline that orchestrates multiple LLM providers to optimize code generation quality, latency, and cost in real time.

Attribute	Details
Stack	Python · TypeScript · LangChain · LangGraph · Ollama · LM Studio · FastAPI · SQLite
LLMs Benchmarked	GPT-4o-mini · Gemini 2.0 Flash · Llama 3.1 8B (via Groq)
Scale	Local + Cloud hybrid; designed for solo developers and small engineering teams
Performance	Zero-token intent classification; provider-aware routing reduces average latency by routing lightweight tasks to local models
Security	Sandboxed code execution; isolated tool registry; environment-scoped secrets via REXODE_ prefix
Impact	Full CLI rebranding (~/.rexode); interactive TUI model picker; multi-agent mode; persistent project memory
Repository

Architecture highlights: The MAQE pipeline operates across six sequential layers — Intent Classification, Provider Selection, Execution, Quality Scoring, Refinement, and Output Validation. A LangChain-based router dynamically dispatches tasks to the optimal provider based on complexity score, context length, and latency budget. The ReAct agentic loop handles tool invocation, error recovery, and iterative refinement without human-in-the-loop intervention.

🔧 Rexode CLI — Developer Tooling & Local LLM Infrastructure

A production-grade command-line interface extending OpenCode CLI with first-class support for local LLM providers, an interactive model selection TUI, persistent chat history, project memory, and sandboxed code execution. Built for solo developers and open-source contributors who prioritize privacy and local-first AI tooling.

Attribute	Details
Stack	TypeScript · Node.js · Ollama API · LM Studio API · Ink (TUI) · SQLite
Scale	Single-binary CLI; zero-config local provider auto-detection
Performance	Streaming inference; lazy model loading; context-aware file injection
Security	Sandboxed subprocess execution; additive-only architecture; no modification to upstream cloud integrations
Impact	7-part sequential feature build system; full provider abstraction layer; web search tooling integration
Repository

Engineering discipline: All feature development follows a strict additive-only constraint — no modifications to existing commands, cloud provider integrations, or upstream agent logic. This ensures clean rebasing against the upstream fork and maintains a stable, auditable diff surface. Features include: local LLM auto-detection, interactive model picker TUI, chat history persistence, file and directory context injection, web search tooling, sandboxed code execution, multi-agent coordination mode, and project-scoped memory.

VIT-AP University — School of Computer Science and Engineering Aug 2025 – Jun 2026 · Amaravati, Andhra Pradesh

Under the supervision of Dr. Prabha Selvaraj, designed and built Rexode — a novel agentic multi-LLM quality optimization framework representing a full-stack AI engineering contribution from architecture through implementation, benchmarking, and academic documentation.

• Designed a six-layer MAQE pipeline integrating LangChain-based LLM routing, zero-token intent classification, and ReAct agentic execution loops
• Benchmarked GPT-4o-mini, Gemini 2.0 Flash, and Llama 3.1 8B (via Groq) across latency, quality score, and cost dimensions
• Authored a 65+ page capstone report spanning 18 chapters covering system architecture, security, scalability, UX, cost modeling, and ethics
• Developed a draw.io system architecture diagram covering Core Pipeline, 6-Layer MAQE, Tool Registry, Memory System, Safety Layer, and multi-provider LLM integration
• Built Rexode CLI with local LLM provider support (Ollama, LM Studio), TUI model picker, persistent memory, and sandboxed code execution

  

current_focus:
  learning:
    - LangGraph multi-agent orchestration patterns
    - LLMOps: evaluation pipelines, model monitoring, A/B testing
    - Distributed systems design for AI workloads
    - Advanced RAG architectures and vector indexing strategies

  building:
    - Rexode: agentic multi-LLM framework with MAQE pipeline
    - Rexode CLI: local LLM provider tooling for solo developers
    - Open-source developer tooling for AI-assisted workflows

  exploring:
    - Speculative decoding and inference optimization techniques
    - Edge deployment of quantized LLMs (GGUF, GPTQ, AWQ)
    - Multi-modal agents and tool-augmented reasoning
    - AI safety and alignment in production agentic systems

  open_to:
    - AI / ML Engineering positions (full-time, 2026)
    - Agentic AI and LLMOps roles
    - Research engineering collaborations
    - Open-source contributions in AI tooling and developer infrastructure

"The best systems are not those that appear intelligent — they are those that make intelligence measurable, reproducible, and reliable at scale."