AI Lab
Run my work yourself.
31 working demos of the systems I build for clients, organized by industry. Each one links back to the production case study it comes from.
Prefer to start from a business problem? Browse by use case →
No API key needed. Every demo plays recorded results instantly, and the in-browser demos run real models on your device. For live AI runs, plug in your own OpenAI or OpenRouter key: it stays in your browser for this session only, is never stored on any server (there is no server), and every request goes directly from your browser to the provider.
Retrieval & RAG
FlagshipThe 2026 RAG stack, not naive top-k: hybrid retrieval (dense + keyword), reranking and grounded answers with citations. Retrieval quality, not the model, is what makes RAG trustworthy.
Security & Safety
Defense-in-depth for LLM apps, mapped to the OWASP LLM Top 10: prompt-injection defense, PII/PHI redaction and real-time crisis detection. From my safety-critical healthcare work.
Healthcare
FlagshipMirrors my production work as sole ML engineer on a mental-health platform: clinical documentation, semantic search and real-time safety.
SOAP Note Generator
Consultation transcript in, structured clinical note out.
Open demo →Similar Transcript Finder
no key neededSemantic search over consultations: meaning, not keywords.
Open demo →AI Assistant Chatbot
One generic bot, many professions: profiles, tones, recent chats and visible context compaction.
Open demo →Safety & Crisis Detection
no key neededFlags a crisis in milliseconds and escalates only the hard cases to a model.
Open demo →Medical Coding Copilot
Clinical note in, suggested ICD-10 and CPT codes with rationale out.
Open demo →Prior-Authorization Letter
Clinical details in, a payer-ready justification letter out.
Open demo →Discharge Summary
Admission and course details in, a structured discharge summary out.
Open demo →Documents & Language
Grounded answers, summarization, extraction and contract intelligence. Built on the systems behind my 92%-accuracy enterprise legal Q&A.
NLP (on-device)
Core NLP running entirely in your browser: zero-shot text classification and semantic similarity, no key and no upload. The on-device building blocks behind intent routing and RAG.
Finance & Insurance
Explainable, compliant automation for KYC, claims and reporting, with the evaluation discipline regulated buyers expect. Built on my enterprise document-processing and multi-agent reporting work.
BI Report Generator
Raw CSV numbers to an executive narrative with insights.
Open demo →Invoice → JSON
Messy invoice text to accounting-ready structured data.
Open demo →KYC Document Extractor
Onboarding documents in, structured KYC data and risk flags out.
Open demo →Insurance Claim Intake
A claim narrative in, a structured claim record and next steps out.
Open demo →Legal
Contract and research AI that cites its sources and refuses to guess. From my multi-agent legal Q&A at 92% accuracy with multi-stage verification.
Customer Experience
Support automation that deflects the routine and grounds every answer in your policy, plus QA scoring to keep quality high as volume scales.
Lead Scorer + Outreach Writer
ICP score with reasoning, then a personalized email that earns a reply.
Open demo →Policy-Grounded Support Reply
Customer message plus your policy in, an accurate on-brand reply out.
Open demo →E-commerce Ops Agent
Ask in plain English about stock and orders; the agent checks the live data and takes the next action, safely.
Open demo →Support Conversation QA
A support transcript in, a scored quality card with coaching out.
Open demo →Voice Support Agent
no key neededSpeak to it and it speaks back, a phone-style agent, right in your browser.
Open demo →Computer Vision
Detection and OCR running fully in your browser, no API key and no upload to any server. From my smart-surveillance research (30+ FPS on edge devices).
Zero-Shot Image Classification (CLIP)
no key neededType your own labels and CLIP scores the photo against them, no training, in your browser.
Open demo →Image Understanding: on-device vs large model
no key neededSee a photo two ways: a light classifier on your device, and a large vision model with your key.
Open demo →Object Detection
no key neededDetect objects in any photo, fully in your browser.
Open demo →OCR, Image to Text
no key neededRead text from photos and scans, in-browser.
Open demo →MLOps & Cost
The production economics behind the models: VRAM sizing, quantization and self-host-vs-API break-even. The 20% that gets attention, and the 80% that decides the bill.