██╗ ██╗██╗██╗ ██╗██╗ ██╗██╗ ██╗ █████╗ ████████╗
██║ ██║██║██║ ██╔╝██║ ██║╚██╗ ██╔╝██╔══██╗╚══██╔══╝
██║ ██║██║█████╔╝ ███████║ ╚████╔╝ ███████║ ██║
╚██╗ ██╔╝██║██╔═██╗ ██╔══██║ ╚██╔╝ ██╔══██║ ██║
╚████╔╝ ██║██║ ██╗██║ ██║ ██║ ██║ ██║ ██║
╚═══╝ ╚═╝╚═╝ ╚═╝╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝
vikhyat = {
"role": "AI Engineer & Builder",
"school": "CSE @ IIIT Sri City",
"focus": ["Agentic Systems", "RAG Pipelines", "MCP", "Static Analysis"],
"ethos": "Ship fast. Stay honest. Let the demo speak.",
}I don't just prototype — I architect systems with real APIs, real constraints, and measurable outcomes.
AI-assisted development is my superpower: I direct the architecture, validate every decision, own the result.
|
Repos → evidence-backed code graphs Local MCP server using Tree-sitter + Git context to map any codebase — dependency tracing, call-flow exploration, blast-radius analysis, and architecture visualization. No execution of the target repo required.
|
RAG-based production support pipeline 5-stage pipeline: safety screening → domain classification → semantic retrieval → grounded reasoning → structured output. Built in 24 hours — six iterations, zero hallucinations in final version, $0 API cost. #153 / 1,349 · Top 11% · HackerRank Orchestrate · |
|
Multi-agent travel disruption → resolved Specialized LangGraph agents handle rebooking, hotel coordination, and proactive passenger SMS — end to end. Reduced rebooking latency from ~45 minutes to under 90 seconds via real-time AviationStack integration. Top 10 @ IdeaVerse'26 · |
SAR + optical → per-building structural risk Dual-branch ResNet18 CNN fusing SAR and optical satellite imagery to produce building-level height estimates on SpaceNet-6. Built solo in 36 hours. MAE 4.48m · R² 0.80 · 1,984 buildings · |
|
7-stage multimodal WhatsApp classifier Classifies messages (notify/digest/mute) using LLMs, speech, and image models over 400+ historical messages. Deterministic scam/phishing/injection detectors short-circuit 22% of messages before any LLM call. #376 / 1,983 · Top 19% · HackerRank Orchestrate · |
Production-grade Python static analysis CLI Deterministic AST-based maintainability metrics with an optional AI reviewer for narrative feedback. Strict modular separation, PEP 561 typed, 85%+ pytest coverage, GitHub Actions CI, Apache 2.0.
|
CDLI Framework · Python · GSoC 2026 Aspirant
- MR 1179 merged — fixed Python 3.12+ PEP 668
externally-managed-environmentinstallation failure; unblocked onboarding for new contributors - 3 additional PRs improving documentation clarity and reducing setup friction through maintainer review
CDLI MCP Server · Python · MCP PoC
- MCP server exposing the CDLI cuneiform artifact database as structured, AI-callable tools (artifact search, metadata retrieval, export)
- Rate limiting, structured error codes, batch request handling, and retry logic for reliable LLM tool-calling
AGENT LAYER LangGraph · MCP · Multi-agent orchestration · Tool-calling
LLM APIs Groq · Anthropic · Gemini · Ollama (local)
RETRIEVAL ChromaDB · MiniLM · TF-IDF · RAG pipeline design
BACKEND FastAPI · Flask · REST APIs · PostgreSQL · SQLite
STATIC ANAL. Tree-sitter · AST Parsing · Python CFG analysis
ML PyTorch · ResNet · CNN · SAR/optical sensor fusion
FRONTEND React · Tailwind CSS · Streamlit
INFRA Docker · GitHub Actions · Git · Linux · pytest
LANGUAGES Python · C/C++ · JavaScript · HTML/CSS
Open to: Remote AI/agent engineering roles · Research collaborations · Open source