Hey, I’m @moonman369
Full‑stack & Web3 roots; currently deep in the lab making large language models do useful things (ideally on purpose). I like turning vague AI ideas into shipped, observable, low‑latency, “wow-that-actually-works” systems.
- 🤖 AI & GenAI – prompt alchemy → hardened pipelines (evals, guardrails, cost + latency dashboards).
- 🧩 AI Agents – planner / executor loops, tool orchestration, memory layers, self‑reflection (minus the existential crisis).
- 📚 RAG – hybrid retrieval (vector + keyword + rerank), smarter chunking, quality metrics > vibes.
- 🛰 MCP Architecture – experimenting with Model Context Protocol as a clean contract between tools, data and agents.
- 🛠 AI-Guided Dev – spec-to-code, test synthesis, refactor assists, documentation autowriting (yes, I let the model explain my own code sometimes).
- 🔁 Feedback Loops – tracing prompts, grounding scores, hallucination audits, & tightening the iteration cycle.
- 💼 Software Engineer @ Tata Consultancy Services (Azure, API Integration, Spring Boot Microservices, .NET Framework).
- 🧱 Blockchain Dev Intern @ W3Dev Private Limited; built & shipped smart contract components.
- 🏁 Chainlink Spring Hackathon 2023 participant—learned fast, shipped faster.
- 🌐 Delivered multiple full stack apps end‑to‑end (design → deploy → post‑mortem / polish).
- 🔗 Crypto playground across: Ethereum, Polygon, Avalanche, Solana, Near, Aptos, Cosmos, Cardano.
- 🧪 Tooling comfort: Hardhat, Truffle, Metamask, Ethers.js, Web3.js, NearJS, etc.
- 🧠 RAG‑Based Code Understanding: Building a retrieval layer that treats source code as a living knowledge graph—hybrid embeddings + structure-aware parsing (AST, symbols, dependencies) → precise, context-packed answers instead of fuzzy guesswork.
- 🏷 Intelligent Prompt Classification: Auto-routing user intent (debug? architectural advice? refactor? explanation?) to tailored system prompts, tools and safety rails—lower latency, higher signal, fewer wasted tokens.
- 🧾 Code Summarization Tool: Multi-resolution summaries (function → module → service) with freshness tracking + diff-aware regeneration; powering “what changed” briefs and onboarding snapshots.
- 🛰 MCP Server Architecture: Designing a modular Model Context Protocol server that exposes tools, datasets, retrieval endpoints and execution sandboxes as composable, observable capabilities for agents.
- Adaptive tool routing (semantic + cost‑aware).
- Vector DB benchmarking (recall vs MIPS vs memory footprint).
- Conversation memory blends (episodic snapshots + semantic embeddings).
- Execution sandboxes for semi‑autonomous agents (security > chaos).
- Web3 × AI intersections (verifiable inference, oracle‑enhanced reasoning).
- Knowledge graphs + embeddings alignment for richer entity grounding.
- Multi‑modal context (logs + code + docs fused for dev copilots).
Ship small. Observe everything. Optimize hot paths. Document intent (so Future Me owes Present Me fewer apologies).
- Core Languages: Python, JavaScript/TypeScript, Go
- LLM Providers: OpenAI, Azure OpenAI, (exploring Anthropic / HF)
- Retrieval: FAISS, pgvector, hybrid BM25 + dense + rerank strategies
- RAG Craft: chunk heuristics, semantic filters, Hallucination vs Grounding eval sets
- Agents: planner–executor loops, tool routing, scratchpad refinement, structured outputs
- MCP: experimenting with Model Context Protocol as a clean tool/data contract
- Observability: prompt traces, latency histograms, cost per feature, eval dashboards
- Dev Assist: AI-driven test generation, refactor suggestions, auto‑doc pipelines
- Frontend: React, Next.js, Vue, Tailwind, HTML5, CSS3
- Backend: Node.js (Express), REST design, microservices (Spring Boot), .NET Framework
- Cloud: Azure (identity, services, deployment flows)
- Data: MongoDB, SQL fundamentals, schema iteration
- Smart Contracts: Solidity, Hardhat, Truffle
- Chains: Ethereum, Polygon, Avalanche, Solana, Near, Cosmos, Cardano, Aptos
- Tooling: Ethers.js, Web3.js, Metamask, Chainlink Oracles
- Patterns: tokens, events, verifying data integrity
- Java, C++, Golang, C#
- Algorithmic reps: LeetCode, HackerRank, Codewars
- Core CS: data structures, OOP, concurrency basics
