Work that ships.

The problem

Startups and businesses in Pakistan had no reliable way to get legally grounded guidance. Generic LLMs gave generic answers with US/UK-style assumptions that didn't apply locally. Legal consultants were expensive and slow.

What we built

PakLawAssist - a RAG-first legal platform grounded to Pakistan's actual legal and regulatory framework. Before generating anything, the system retrieves relevant local legal context. Contract drafting, contract review, compliance guidance, and petition drafting are grounded to Pakistan-specific law, not generic templates.

What's live

• Full application at agent.paklawassist.com
• Knowledge chatbot deployed with SMEDA
• Knowledge chatbot deployed with PSIC
• International version live at nuworo.com

Technical depth

Multi-step RAG orchestration: retrieve, reason, generate, validate, and format. Role-based access control for institutional deployments. Usage quotas and admin governance. Clean document outputs - not chat text.

The problem

Doctors were spending a disproportionate amount of every consultation on documentation - typing while patients waited. It wasn't just an efficiency problem. It was a fatigue problem. And a patient experience problem.

What we built

NurMed - an AI clinical documentation platform that handles the typing so doctors don't have to. Structured note generation, clinical summaries, and consistent output format. Built to work in a live clinical environment, not a controlled demo.

The problem

Star ratings are blunt. A 4.2 doesn't tell you what guests loved, what's breaking the experience, or what changed last month. Hotel and restaurant operators were making decisions based on averages - and missing everything specific and actionable underneath them.

What we built

Senteez - a feedback intelligence platform that ingests reviews from Google Business, TripAdvisor, Facebook, BestBuy, and more, runs NLP analysis with fine-tuned models, and extracts structured insights. Sentiment, intent detection, topic modeling, and named entity recognition turn reviews into a decision engine.

What's live

• 11,127 reviews across 8 Dubai Marina hotels
• 6,007 reviews across 9 Dubai Marina restaurants
• 6,067 reviews across 5 US coffee brands
• Live at senteez.com

The problem

Writers - from students to professional authors - were spending weeks or months getting to a first draft. The blank page problem is real. Most AI writing tools give you a paragraph, maybe a scene. Nobody had solved the full manuscript.

What we built

FictionPub.ai - an enterprise-grade manuscript generation platform that takes a writer from concept to complete novel. The pipeline covers market research, story concept, character development, chapter-by-chapter plot structure, full draft generation, and quality evaluation. Every section is assessed by an LLM-as-Judge system before delivery - creativity, coherence, engagement, and pacing. Below 70%, it regenerates automatically.

What's live

• Live at app.fictionpub.ai

The engineering behind it

• Multi-provider AI strategy - Anthropic + OpenAI, model selected per task type
• Celery + Redis distributed task queue - handles 20-30 minute generation jobs at scale
• WebSocket real-time streaming - writers see content generated paragraph by paragraph
• LLM-as-Judge quality gates - automated evaluation and regeneration before delivery
• Opik prompt management - 500+ genre-specific prompts with versioning and A/B testing
• Auto-recovery - stuck task detection, resume from any stage, zero lost work