|
I'm an AI developer and vibe coder. I take an idea and ship it as a working AI product, fast, with LLMs, agents and AI coding tools doing a lot of the heavy lifting. My focus is prompt engineering, multi-agent systems and full-stack LLM apps. Right now I'm working on a batch of new projects. I'm based in London, where I'm also studying IT. ⚡ Currently
|
|
LLM apps, agents and developer tools, built fast and shipped in public.
|
|
LLM-powered apps and multi-agent systems: planners, researchers and writers that hand work to each other and stay traceable. agent-relay · lexcore |
Prompts as versioned, testable code, scored against rubrics so every change is measured before it ships. prompt-forge |
Idea to deployed app, fast. AI pair-programming for the boilerplate, my judgement on what actually ships. TypeScript · React · Python |
Local-first clip studio: cuts pauses, ranks highlights, burns animated word-by-word captions and batch-exports Shorts, TikTok, Reels and 16:9. Live demo. TypeScript React ffmpeg Whisper
|
Lightweight multi-agent orchestration: planner, researcher and writer agents relaying typed messages over a traceable bus. TypeScript Node.js Agents
|
Versioned prompt templates and a rubric-based eval harness, so you know v2 beats v1 before it ships. Offline-first. Python LLM evals CI
|
AI legal guidance assistant: describe a legal issue, get plain-English rights and next steps. Cinematic, animated UI. TypeScript React Groq · Llama 3.1 tRPC
|
Dependency-free LLM safety toolkit for prompt-injection detection, reversible PII redaction and output policy checks. Python LLM security Guardrails
|
Long-term memory for AI agents with episodic and semantic stores, TF-IDF recall and decay/importance scoring. TypeScript Agents TF-IDF
|
Offline-first RAG toolkit with BM25, hybrid retrieval, cited answers and evaluation gates. Python RAG BM25
|
A character-level GPT in pure NumPy with handwritten backprop, gradient checks, training and sampling. Python NumPy Transformers
|
Polished streaming LLM chat UI with a mock backend, Markdown rendering and local history. React Vite TypeScript
|
Offline-first diff reviewer with 15 rule checks, optional LLM notes and a GitHub Action. TypeScript GitHub Actions Static analysis
|
Sales and footfall forecasting with seasonal-naive, Holt-Winters and gradient-boosting backtests. Python scikit-learn Forecasting
|
Offline sentiment analysis with lexicon and trainable models, incident alerts and a static dashboard. Python NLP Analytics
|
Classical computer vision for scene tags, alt text, colours and self-contained HTML reports. Python Computer vision NumPy
|
Turn meeting transcripts into owners, due dates, decisions and concise Markdown notes. Python Speech-to-text Productivity
|
Explainable local-business lead scoring with weighted and logistic-regression models plus ranked reports. Python Sales Explainable AI
|
Explainable phishing email & URL detector with feature reasons and a CLI. Python Cybersecurity ML
|
Security log anomaly detection for SSH/auth and web logs with rules, Isolation Forest and MITRE-mapped reports. Python Blue-team Anomaly detection
|
Defensive project auditor for secrets, vulnerable deps and Dockerfile/CI misconfigs; text, JSON or SARIF. Python DevSecOps SARIF
|
YAML AI automation workflows with LLM, HTTP, webhook and email-draft steps, mock runner and CLI. Python Agents YAML
|
Denoising diffusion (DDPM/DDIM) from scratch in PyTorch on 2D toy datasets, with denoising GIFs. PyTorch Diffusion From-scratch
|
Convolutional VAE from scratch in PyTorch on synthetic digits, with latent-space visualisations. PyTorch VAE Generative
|
More projects
- scene-director: screenplay → scene-by-scene AI shot list with ready-to-paste image/video prompts.
Python - local-biz-template: fast, SEO-ready site template for local businesses, driven by one JSON file.
JavaScriptTailwind - hasenix-portfolio: animated single-page portfolio site, no framework.
HTMLCSSJS - ugc-script-gen: product brief → ranked, shot-ready UGC ad scripts and hooks with explainable scoring.
PythonLLM
| 01 · Ship first, polish fast. Get a working version in front of people, then iterate in tight loops. |
02 · AI is my pair programmer. It writes the boilerplate. I own the architecture, the review and the result. |
| 03 · Prompts are code. Versioned, diffed and tested. If I can't measure it, I don't trust it. |
04 · Small agents, clear contracts. Typed messages, readable logs, replayable runs. No black boxes. |
| 05 · Vibes, then verification. Move fast on feel, then lock it in with tests and evals. |
06 · Build in public. Open-source the useful bits and share what I learn along the way. |
08 LET'S TALK
Building something with AI, or want a fast prototype of an idea?
Drop me a message. I'm always up for talking agents, LLM apps and shipping fast.
Mohammed Hassan bin Tayyeb · AI developer · London, UK · hasenix.com