# The Hallucinated Lab > The Hallucinated Lab is a local-first AI lab and consultancy. It builds AI tools that run entirely on the user's own machine — no cloud dependency, no API keys, no data leaving the device — and provides AI consulting and engineering services to individuals and businesses. It was founded by Pratyush and Divyansh Tripathi, both computer science undergraduates at Manipal University Jaipur, India. The Hallucinated Lab operates in three modes: 1. **Free artifacts.** Prompts, LoRA adapters, open-source packages, interactive explainers, and a browser-based assistant that understands plain-English tool requests. All free, all designed to run offline. 2. **Services.** AI consultancy for individuals (mentoring, code and architecture review, research support) and businesses (feasibility audits, RAG and agent system builds, private on-premise deployment, team enablement). 3. **Certification.** Project-graded course tracks in local-first AI. Every track ends in a build that is reviewed against published criteria before a certificate is issued. **Core position:** if a model can run on your hardware, you should be able to run it without a subscription, a rate limit, or a third party reading your data. Every tool published here is built to that constraint. **Site is:** a static, zero-build site (vanilla HTML, CSS, JavaScript) hosted on GitHub Pages. No frameworks, no trackers, no analytics scripts, no cookies. ## Pages - [Home](https://thehallucinatedlab.space/): Overview of the lab and a guide to every section of the site. - [Tools](https://thehallucinatedlab.space/tools.html): Gateway to the four tools — Convert, Assistant, Prompts and LoRA adapters. Each has its own page. - [Convert](https://thehallucinatedlab.space/convert.html): Converts images between PNG, JPEG, WebP and AVIF entirely in the browser on a canvas. Nothing is uploaded and it works offline. Doubles as the argument reference for the convert method in the thehallucinatedlab pip package. - [Assistant](https://thehallucinatedlab.space/interface.html): Browser interface that parses a plain-English request and runs the matching tool locally. No model, no install, no cloud relay. - [THL Library](https://thehallucinatedlab.space/library.html): The full package shelf - pip-installable packages, embeddable engines and open-source building blocks, with install and usage for each. - [Solutions](https://thehallucinatedlab.space/solutions.html): Shipped products, led by ScoobyBench, a Windows desktop app that benchmarks real AI inference performance against vendor claims. - [Media](https://thehallucinatedlab.space/media.html): Entry point to Blogs, Artifacts, and (upcoming) Notebooks. - [Certification](https://thehallucinatedlab.space/certification.html): Six project-graded course tracks and the criteria a capstone must clear. - [Sitemap](https://thehallucinatedlab.space/sitemap.html): Every page on the site in one list, grouped by section, with a line on what each is for. - [Consultancy](https://thehallucinatedlab.space/consultancy.html): Engagement models for individuals and businesses, the four-stage process, and how to start. ## Tools - [Prompts](https://thehallucinatedlab.space/prompts.html): Eight model-agnostic prompts for code review, root-cause debugging, refactoring, dataset triage, SQL generation, tutoring, paper review, and model routing. Free to copy, no account. - [LoRA adapters](https://thehallucinatedlab.space/adapters.html): Small fine-tuned adapters that stack on a base model already installed locally. Currently in training; none are downloadable yet. Usage guides for Ollama and PEFT are published. - [Convert](https://thehallucinatedlab.space/convert.html): The first THL tool. Converts PNG, JPEG, WebP and AVIF client-side using the canvas API - no upload, no server, works offline. Arguments: format (required), quality (1-100, default 92), background (hex, flattens transparency for JPEG). - [NexusLink Engine](https://thehallucinatedlab.space/solutions.html): Headless Rust SDK for encrypted peer-to-peer LAN/WAN communication. Built from source rather than PyPI for now. - Python toolkit: `pip install thehallucinatedlab` gives every tool as a Python method plus a `thl` command. ## Media - [Blogs](https://thehallucinatedlab.space/blogs.html): Long-form writing on AI, quantum computing, privacy, and developer tooling. Includes a community section open to reader submissions. - [Artifacts](https://thehallucinatedlab.space/artifacts.html): Interactive explainers the reader operates rather than reads. All computation runs in the browser. - [Time & Space Complexity](https://thehallucinatedlab.space/blogs/complexity.html): Interactive teaching session on asymptotic analysis, Big-O through the Master Theorem, with a growth-rate explorer and twelve worked examples. - [The Model Is No Longer the Product](https://thehallucinatedlab.space/blogs/ai-orchestration.html): Interactive piece on AI engineering as orchestration, with a runnable RAG pipeline and a failure-diagnosis game. - [The Future of Local-First AI](https://thehallucinatedlab.space/blogs/sample-blog.html): Blog on the shift from cloud dependency to local-first intelligence. ## Services - [Consultancy for individuals](https://thehallucinatedlab.space/consultancy.html#individuals): Project mentoring, code and architecture review, research and experiment design, portfolio work. For students, researchers, and solo builders. - [Consultancy for businesses](https://thehallucinatedlab.space/consultancy.html#businesses): Fixed-scope feasibility audits, RAG and agent system builds, on-premise and air-gapped deployment, private delivery of certification tracks for internal teams. - [Engagement process](https://thehallucinatedlab.space/consultancy.html#process): Free scoping call, written proposal with fixed price where scope allows, build in the open with repo access from day one, handover with documentation and a runbook. ## Technical capabilities - Retrieval and RAG: document pipelines, chunking strategy, hybrid dense and keyword search, reranking, retrieval evaluation separated from generation evaluation. - Fine-tuning: instruction dataset construction, LoRA and QLoRA training on single consumer GPUs, held-out evaluation and contamination checks. - Agents and tooling: tool interface design, routing between small and large models, budgets, timeouts, guardrails, trace logging. - Private deployment: on-premise and air-gapped model serving, hardware sizing, quantization trade-off analysis, benchmarking measured hardware against vendor claims. ## Contact - Email: thehallucinatedlab@gmail.com - GitHub organisation: https://github.com/The-Hallucinated-Lab - Pratyush: https://github.com/06pratyush — AI, machine learning, data science - Divyansh Tripathi: https://github.com/TheQMLGuy — quantum computing, machine learning, developer experience ## Optional - [Full site map for crawlers](https://thehallucinatedlab.space/llms-full.txt): Expanded directory of every page and section with descriptions. - [XML sitemap](https://thehallucinatedlab.space/sitemap.xml) - [Source repository](https://github.com/The-Hallucinated-Lab/thehallucinatedlab)