[ SYSTEM INITIALIZATION ]
> LOADING MODULES... OK
> ESTABLISHING CONNECTION... OK
>

Swayam R. Gavankar

> Systems, Backend & Agentic Systems Engineer | Specializing in Rust, Go, AI Agents & MCPs
[ SCROLL DOWN TO CONTINUE ]

[ NARRATIVE_PROFILE ]

I am Swayam R. Gavankar, a systems and backend engineer specializing in low-level systems, agentic software engineering, and modern development workflows. I possess deep experience utilizing Git, GitHub collaborative pipelines, and CI/CD tools to deliver robust, high-performance production systems while eliminating architectural overhead. I excel at creating and orchestrating autonomous LLM agents, designing multi-agent routing systems, and developing custom Model Context Protocol (MCP) integrations that bridge models with external environments. In my daily development, I pioneer disciplined agentic workflows, writing structured blueprints and plan documents for AI agents to execute. This structured process guarantees clean, deterministic, and bug-free codebases. As an adaptive, self-starting systems programmer, I rapidly master new stacks and build high-performance systems from the ground up.

[ SKILLS & CAPABILITIES ]

> PROGRAMMING LANGUAGES
[ Python ][ Rust ][ Go ][ TypeScript ][ JavaScript ][ C++ ]
> SYSTEMS, BACKEND & OPS
[ FastAPI, Django, Django Rest Framework, Flask ][ React, NextJS, Express ][ Git & GitHub ][ gRPC & High-Performance Communications ][ Microservices & System Design ][ PostgreSQL, SQLite, Redis ][ Docker & Containerization ][ Debian Server Administration & CI/CD ]
> AGENTIC SYSTEMS, AI & MCP
[ Supervised Fine-Tuning (SFT) & PEFT (LoRA, QLoRA) ][ Model Context Protocol (MCP) Server Development ][ Local SLM Deployments (llama.cpp, LiteRT) ][ HuggingFace Transformers & Tokenizers ][ Creating Custom AI Agents & Multi-Agent Routers ][ Structured Output Tuning (JSON / YAML / TypeChat) ][ Data Analysis: Pandas, NumPy ][ Tool Calling, Function Calling, & Tool Integrations ][ Agentic Workflows (Structured Blueprints & Plan Execution) ][ High-Performance Code Optimization & Debugging ]
> SPECIAL SKILLS
[ Japanese: JLPT N3 Certified ][ Languages: English, Marathi, Hindi ]
> EXECUTION STYLE
  • Strict Documented Agentic Planning: Authoring precise planning blueprints and markdown steps for AI agents, optimizing context windows, token usage, and preventing hallucinations.
  • Low-Level Manual Debugging: Resolving obscure runtime issues and race conditions using system diagnostics, traces, and memory profiles.
  • Modern Git-Driven Development: Leveraging version control best practices, continuous integration, and automated test runners to maintain clean, production-ready repositories.

[ PAST_PROJECTS ]

>Stric: High-Performance QUIC & Flow Mesh Networking(Open Source)

Authored a high-performance network transport and mesh routing ecosystem in Rust built on QUIC. Includes `stric-core` (symmetric P2P wrapper over `quinn`), `stric-tower` (an `axum`-like request-response framework), and `stric-flow` (topic-based mesh routing and state sync engine using Dijkstra calculations). Maintained using a rigorous Git branch-and-PR workflow with GitHub Actions for automated unit and integration tests.

#Rust#QUIC#Asynchronous#P2P Mesh#Tower#Dijkstra#Git#GitHub Actions
>env-doctor: ML Runtime Compatibility & Intelligence Platform(Open Source (Hackathon))

Designed and engineered a local-first runtime compatibility and intelligence platform for the HuggingFace, PyTorch, and CUDA ecosystem. Exposes Model Context Protocol (MCP) server endpoints to verify software/hardware ABI configurations, execute smart VRAM estimation, and run precise OOM (Out-Of-Memory) checks for vLLM, DeepSpeed, and llama.cpp runtimes. Features Watsonx Orchestrate integration to automatically process compatibility reports, leveraging an MCP agent to write and commit rules to a community compatibility database.

#Python#PyTorch#MCP#CUDA#Model Context Protocol#Watsonx#vLLM#DeepSpeed
>Mandala Workspace: Distributed Hierarchical Filesystem(Open Source)

Engineered a secure, distributed, automatically-versioned hierarchical monorepo workspace. Implemented challenge-response authentication (Ed25519), PASETO session management, local keyring encryption (AES-256-GCM, Argon2id), custom folder access control bitmasks, and hierarchical cache invalidation via Redis. Managed via Git with customized GitHub pre-commit hooks to automate linting and security scans.

#Go#gRPC#Wails v2#React#PostgreSQL#Redis#Git#Security Hooks
>Student Buddy: Privacy-First AI Academic Advisor (Multi-Agent System)(Open Source)

Created a local-first, privacy-focused academic advisor by developing and orchestrating a multi-agent routing system. Built specialized, autonomous agents (Policy, Academic, Advisor, and Fusion agents) using state machines, custom tool calling, and local SLMs via llama.cpp. Implemented custom prompt chains and structured outputs.

#FastAPI#Next.js#llama.cpp#Gradio#Multi-Agent Routing#AI Agent Creation
>Proprietary Options Trading Microservice(FinTech Client (Proprietary))[ CONFIDENTIAL ]

Engineered a small but efficient proprietary options trading microservice in Python using Pandas, NumPy, and FastAPI to process trading orders and option chains in milliseconds. System processes millions of trading symbols and 1000s of options concurrently, executes custom business logic, manages real-time active trade states, and handles automatic exit signals via trading APIs. Developed using collaborative Git workflows with detailed pull request code reviews.

#Python#FastAPI#Pandas#NumPy#Concurrency#Git Workflow
>Blogistan: Full-Featured Blogging & Interaction Engine(Open Source)

Built a server-side rendered blogging platform using Django and Django Rest Framework. Programmed an asynchronous, client-side commenting and reply system leveraging browser fetch APIs and CSRF validation, integrated with locally-bundled Summernote WYSIWYG editor. Collaborated with contributors using Git and GitHub issues.

#Django#Python#SQLite#REST API#JavaScript#Summernote#Git