Full-stack engineer who pairs production React/TypeScript front-ends with Python AI services in real shipping products. Two years freelancing 20+ LLM-powered apps for paying clients on Claude, OpenAI and LangChain. MSc Advanced Data Science & AI at the University of Liverpool.

I write code that other people maintain, document the workflows so non-engineers can run them, and use AI tooling daily as part of the work rather than as a side project.

const vaibhav = {
  role:       "Full-Stack + AI Engineer",
  location:   "Liverpool, UK",
  status:     "UK Right to Work · Graduate visa from Jan 2027",
  stack:      ["TypeScript", "Python", "React 19", "Next.js 15", "FastAPI"],
  data:       ["Postgres", "Supabase", "pgvector", "Redis", "BullMQ"],
  ai:         ["LangChain", "OpenAI", "Gemini", "Whisper STT", "MCP servers"],
  shipping:   ["DRIP (AI sneaker stylist · multi-agent AI)", "DelOS (Agents Under Pressure '26)", "Qyntra (Wikithon '26 Finalist)", "MCP Marketplace", "Scholar.AI v2"],
  open_to:    ["AI / Full-stack roles", "Graduate placement Jan 2027"],
};

🏆 HydraSentry  ·  live  ·  HydraDB Build Blitz · Winner
_{Graph-native memory-integrity layer for AI agents. One poisoned memory can silently override an agent's policy on every future run, invisible to prompt firewalls because the poison is in what the agent retrieves, not the prompt. HydraSentry replays the agent on clean vs poisoned context, traces the exact taint path through the real HydraDB query_paths graph, blocks the unsafe action via an MCP firewall before the agent acts, and seals a signed, offline-verifiable Memory Integrity Certificate. Multi-tenant, semantic paraphrase detection via embeddings, self-verifying against the OWASP Agentic Top 10. Live, no login.}

🤖 Kodro  ·  Offline robot design and simulation studio · MSc dissertation
_{Design a robot, program it, watch it work, fully offline with no account and no cloud. Build a custom machine in the Robot Lab, program it in a Python subset, blocks, or with a local AI assistant (Ollama, with a rule based fallback), then validate it in a 3D world picked to fit it: a self driving car gets a City with looping traffic and a crossing, a home or arm robot gets a Room, a rover gets planetary terrain. Motion reads per type with weight transfer and suspension. A localStorage memory reflects on each run and keeps saved skills, so it refines from its own verified work without retraining any weights. Vendored React and Three.js, pre compiled offline. Verified: 21 of 21 interpreter QA and 851 engine tests at ~86% coverage.}

👟 DRIP — AI Sneaker Stylist  ·  live  ·  Multi-agent AI · Live StockX data
_{Builds outfits around your sneaker collection. A Stylist agent drafts, a Critic agent refines — honest weather/material advice (it tells you not to wear suede in the rain). Live StockX resale prices via KicksDB, camera color-scan with auto-capture, a real recommendation engine, live Reddit/YouTube/news feed. React 19 · Vercel serverless · NVIDIA / OpenRouter / HuggingFace.}

🏆 Qyntra  ·  live  ·  Wikithon '26 Finalist
_{Your own private Wikipedia, compiled from your Notion, Drive, Gmail, GitHub and desktop files. Cited answers. 3D galaxy view. Groq Llama 3.3 70B. Built solo in a day with OpenCode Go + Claude Opus 4.7.}

🧠 Continuity  ·  live  ·  Agents You Love Hackathon · Best Use of Memory
_{Most agents forget the moment a tab closes. Continuity remembers what matters, lets the rest decay, supersedes itself when you contradict yourself, and closes your open loops before you ask. HydraDB is the primary memory: every fact is a real write and every recall a real query, so your context survives a reload with nothing in localStorage. Say a doctor wants to prescribe amoxicillin and it recalls your stored penicillin allergy and warns you, the answer changing because of memory. One memory across a web app, an MCP server that follows you between Claude Code, Codex, Cursor and Gemini CLI, and a 24/7 daemon, with an execution log proving every HydraDB write and query.}

💸 Recoup  ·  live  ·  Google Cloud Rapid Agent Hackathon · MongoDB track
_{Budgeting apps show where your money went. Recoup gets it back. One Google sign-in scans your real Gmail receipts read-only, then a fleet of agents grounds every claim in consumer-protection law via MongoDB Atlas Vector Search, drafts it with Gemini 3 through the Google ADK, runs an independent verifier, and stops at a human approval gate. One approval and a real Playwright browser walks the vendor cancellation portal and streams a live screenshot back. A Sentinel watches your drains on a schedule, and it searches 20,000 real California unclaimed-property records. Every step writes to a tamper-evident sha256 hash chain. Walkthrough.}

🛡️ RecallOps Cortex  ·  live  ·  Google Cloud Rapid Agent Hackathon · Fivetran track
_{Turns a real openFDA food recall into a scoped blast radius across Fivetran-synced BigQuery ops data, has Gemini draft six containment actions (each citing its evidence), gates every action behind a human approval, and writes each step to a tamper-evident sha256 hash-chain audit log. FastAPI on Cloud Run, a 14-route command center, honest live vs fallback badging. Whitepaper.}

🛠️ DelOS  ·  live  ·  Agents Under Pressure '26 · AIValley × HydraDB
_{Browser-OS where multi-agent orchestration is wired in by default. xterm.js Agent Fleet running parallel agents on isolated HydraDB tenant subnamespaces (serverless git-worktree). Force-directed memory graph over HydraDB graph+vector. 6-provider LLM cascade (Groq + Mistral + Gemini + NIM + Cerebras + Bytez + OpenRouter). 79 endpoints. 14/14 regression + 62/62 brutal probe pass. Built solo across a 48h sprint.}

MCP Marketplace · live Registry of 800+ MCP servers.	Scholar.AI AI paper assistant. Chrome MV3.	MedReviewAI · live PICO extraction across PubMed.
CogniLoop · live Socratic AI tutor. Hints, not answers.	Scoping Review RAG LangChain pipeline for systematic reviews.	EdgeBench Local Ollama LLM benchmark + guardrails.
Portfolio · live Hand-coded personal site.	Antivibe Claude Code skill. Learn what AI writes.	View all repos → 20+ public projects.

_Languages

_Frontend

_{Backend & Data}

_{AI / LLM}

_{DevOps & Tooling}

• 🛠️ Agents Under Pressure '26 — shipped DelOS · browser-OS for multi-agent work · xterm Agent Fleet · HydraDB graph+vector · 6-provider cascade
• 🏆 Wikithon '26 Finalist with Qyntra · personal Wikipedia for your files
• Shipping MCP Marketplace. Public registry, auto-syncs daily · mcp-hub-registry.vercel.app
• Iterating on Scholar.AI v2. Gemini PDF, side-panel chat, library mode
• MSc Advanced Data Science & AI at University of Liverpool (Jan 2026 to Jan 2027)
• Reading: agentic eval frameworks, prompt-cache patterns, MCP server design
• Shipped Own Wiki (mnemosyne), a local-first knowledge OS in one portable Windows exe
• Top 3, Hack HydraDB security hackathon, responsibly disclosed an authentication weakness, acknowledged by the maintainers
• Published NEXUS, a preprint on cloud gaming latency, Zenodo DOI

_{github.com/vaibhav4046 · 2026}