01
Agentic AI for Regulated Industries
Compliance-grade AI for pharma, medtech and financial services — where an answer that sounds right isn't good enough. Traceable evidence, auditable reasoning, and accuracy you can defend to a regulator.
AI ProductArchitect.
I design and build agentic AI systems for regulated industries — pharma, medtech and financial services.
This work used to take four people: a product manager, a solutions architect, an engineer and a consultant. I hold the business problem and the technical system at the same time, and own the result from strategy through to production.
01$1M+Revenue delivered
02100%Job success score
0320+AI products shipped
049+Years in product
01Profile
From vision to launch.
Founder of Zaltech AI, an AI product studio delivering across healthcare, legal tech, real estate and enterprise B2B.
I'm Sohaib Waheed, an AI Product Architect and the founder of Zaltech AI — a studio I've scaled to a team of 20+ and more than 20 shipped products.
Most of my work now sits in regulated industries, where an answer that merely sounds right isn't good enough. For a medtech client, I architected a multi-agent compliance engine that took regulatory document accuracy from 41% to over 90% across FDA 510(k) submissions.
What I do isn't consulting in the usual sense. I sit with a business, find where AI creates real value rather than where it sounds impressive, design the architecture, and then build it — retrieval pipelines, multi-agent orchestration, voice systems, and the evaluation harnesses that keep them honest in production.
$1M+ in revenue delivered across 101 client engagements, at a 100% Job Success score.
Expert-VettedTop 1% on Upwork
ToptalVetted network
100%Job success score
UTS SydneyInformation Systems, with distinction
Products we ownZaltech AI
01
Zagent AI
One prompt to a live business
02
MedScribe AI
Clinical documentation
03
Zal GPT
Private enterprise intelligence
04
Levo AI
AI voice agents for sales
05
PropertyPilot AI
Real estate automation
06
SalesAssist AI
E-commerce sales assistant
07
KnowledgeBase AI
RAG-powered support
White-label platforms I built and license, alongside 100+ client engagements delivered through the studio.
02Focus Areas
Where I focus.
Two sides of the same work: deep AI systems architecture, and the product judgment to turn that capability into something people actually adopt.
Agentic Systems
Multi-Agent Architecture
Reusable infrastructure, so the second workflow costs a fraction of the first.
- Autonomous task decomposition, planning and execution.
- Long-term memory that doesn't drift between sessions.
- Hallucination and cost tracking across every agent.
- Tool-calling into the APIs and databases you already run.
03Signature Architecture
Memory is the moat.
Most AI products stop at retrieval. The systems I design layer semantic, relational, temporal and behavioural memory so context compounds instead of resetting every session.
Four-layer stackselect a layer
L1 · Vector<50ms retrieval
Semantic Memory
Every communication, approved response and interaction, embedded and searchable by meaning rather than keyword.
- text-embedding-3-large for text, CLIP for multimodal.
- Pulls the twenty closest prior examples on every request.
- Drives few-shot injection, so personalisation needs no per-user fine-tuning.
Context reasoning engine · request flowEvery request assembles context across all four layers in parallel.
01Intent classification
02Parallel query across all four layers
03Relevance ranking: recency, match, diversity
04Prompt assembly from top examples + rules
05Model routing
06Self-critique across candidate drafts
07Confidence gate: send, flag, or escalate
Human stays in the loop wherever confidence is low.
04Capabilities
How I can help.
Engagements are scoped to the problem, not to a package. Every one starts with a discovery call.
02
Multi-Agent Orchestration Architecture
Reusable infrastructure for autonomous task decomposition, planning and execution using LangGraph, LangChain and CrewAI — designed so your second and third workflow cost a fraction of the first.
03
AI Coaching for Founders
One-to-one coaching for founders who want to build with AI themselves. How to scope a product, pick the right models and tools, use AI-assisted development to ship, and tell a real capability apart from a demo — the same playbook I run in my own studio.
04
RAG & Retrieval Architecture
Retrieval systems your team can actually trust, built on Pinecone, Weaviate or ChromaDB — with the evaluation harness and guardrails that keep retrieval quality measurable over time.
05
Voice AI & Conversational Systems
Production voice agents on Vapi, Retell and ElevenLabs handling qualification, scheduling and support. Deployed at over 1,000 calls a day for real estate and healthcare clients.
06
Fractional Head of Product / CPTO
Product leadership for teams building AI without a senior AI product voice in the room — roadmap, discovery, positioning and technical architecture, held by one person accountable for the outcome.
07
AI Integration & Workflow Automation
Connecting AI to the systems you already run — CRMs, ERPs, Slack, internal tooling — via n8n, Make and direct API work, without breaking what currently works.
Drag / scroll06 capabilities
05Trajectory
The path here.
Consulting, then product, then building it myself — each step narrowing toward AI in regulated environments.
Span
2016
2026
Ten years, one direction of travel.
2016
Consulting Foundation
Graduated from UTS Sydney and started in management consulting, learning to frame messy business problems before reaching for a solution — a habit that still shapes how I scope AI work.
2018
Into Product
Moved into product management. Discovery, roadmaps, Agile delivery, and the unglamorous work of finding out what users will actually adopt rather than what sounds good in a review.
2020
The GPT-3 Pivot
Saw GPT-3 and reorganised my career around it. Built early LLM applications and retrieval systems while most of the industry was still treating language models as a novelty.
2021
Product Management in Engineering
Ran product and project management inside an engineering department, working close enough to the build to stop being the person who only writes the ticket.
2022
Architect, Not Just Manager
Started building what I specified. LangChain, vector databases, and AI-assisted development collapsed the gap between the roadmap and a working system.
2023
Founded Zaltech AI
Launched an AI product studio for zero-to-one work, and grew it to a team of 20+ delivering across healthcare, legal tech, real estate and enterprise B2B.
2024
Products at Scale
Shipped a voice AI platform handling over 1,000 calls a day, and launched PropoZal.ai to 100 users and 2,000+ generated proposals within three weeks of release.
2025-26
Regulated Industries
Focused the studio on compliance-grade AI. Architected a multi-agent compliance engine for a medtech client that lifted FDA 510(k) document accuracy from 41% to over 90%.
06Selected Work
Shipped, not slideware.
Systems in production across healthcare, legal, compliance and regulated enterprise.
Our productZaltech AI
Zagent AI
Type one prompt, get a live business. A no-code AI builder for non-technical founders — it plans the pages, generates the screens, scaffolds the API, then deploys in a single click.
1
Prompt to build
1-click
Deploy
Full-stack
Output
Next.jsMulti-Agent PipelinePython APIOne-click Deploy
HealthcarePlayingAI-Powered Medical Wellness Assistant
AI website assistant for a medical wellness clinic offering IV therapy and consultations. Answers patient queries, books appointments directly and pushes live promotion updates.
Real-time
Responses
Calendly
Direct booking
Live
Promo sync
FastAPIGPT-4CalendlyMongoDB
EducationPlayingAdaptive Learning Platform
Fully automated AI assessment platform that removed manual grading, testing across reading, writing, speaking and listening with AI-generated reports for institutions.
4
Skill areas
Automated
Grading
AI
Report generation
FastAPIGPT-4WebSocketsSpeech-to-Text
Real EstatePlayingAI-Powered Homeowner Data Analysis
Compliance intelligence pipeline for property data — ingesting, normalising and analysing records at scale to surface actionable owner and property signals.
500K+
Records processed
Automated
Ingestion
AI
Enrichment
PythonFastAPIPostgreSQLGPT-4
EnterprisePlayingRevolutionizing Proposal Generation with AI
Smart RFP automation that generates detailed 50–60 page proposals in minutes, with section-level editing via adaptive AI, manual input or chat, and live collaboration.
50-60
Pages
Minutes
Generation
Live
Collaboration
FastAPIGPT-4WebSocketsCelery
LegalPlayingAI-Powered Legal Research & Automation
Legal research and drafting platform grounded in Chilean civil law, with retrieval over a corpus of more than 156,000 statutory and case documents.
156K+
Documents
RAG
Retrieval
Admin + User
Portals
FastAPIGPT-4PineconeRAG
HealthcarePlayingRevolutionizing Digital Therapy with AI
Mobile mental health platform with AI therapists across CBT, DBT and other modalities in chat, audio and video, plus symptom, sobriety and wellness tracking.
4
Therapy modes
3
Formats
Mobile
Platform
React NativeFastAPIGPT-4oLangChain
07Method
How I work.
Four phases. The first one ends with a yes or a no rather than a proposal.
Phase 01
Discovery
- Your problem, users and constraints
- Honest feasibility assessment
- Where AI adds value — and where it doesn't
- A straight answer on fit
Phase 02
Architecture & Roadmap
- System and data architecture
- Evaluation criteria defined up front
- Scope, milestones and timeline
- Decisions documented before code
Phase 03
Build & Ship
- Delivery in focused sprints
- Working software reviewed weekly
- Integrations and testing in parallel
- Continuous deployment from day one
Phase 04
Launch & Support
- Production deployment and monitoring
- Handover with real documentation
- 30 days of support included
- Optional retainer for what comes next
08The Paper-First Method
Everything I build starts on paper.
The method behind 25–30+ applications. I do the thinking, the AI does the building. The templates are free.
Paper → Pushback → Production.
AI can write the code in minutes. It can't decide what's worth building. So every build starts with one page, a voice session where Claude has to push back, and a six-line brief, before a single prompt is written.
My glass wall · Oct 2024paperfirstmethod.com ↗
01Step 1
Paper
One page, pen not keyboard. The flow, the journey, the architecture. Where the AI sits, and where a human must say yes.
02Steps 2 – 4
Pushback
I read the page out loud to Claude and make it disagree. Five questions, until the page is honest. Then I decide the architecture and the stack.
03Steps 5 – 7
Production
A six-line brief becomes one build prompt per service. The AI builds inside the structure I decided. I review like an architect, ship, and go back to paper.
09How Zaltech AI runs
Organised around decisions, not people.
For a hundred years, more work meant more people. AI broke that link. So I stopped drawing my org chart around job titles and started drawing it around the calls a human has to make.
Sohaib Waheed
Sets the direction
Head of Product
What to build
A human decidesArchitect
GitHub · Linear
Designer
Figma · Vercel
QA Manager
What ships
A human decidesReviewer
GitHub
QA Tester
Vercel · Postgres
Finance Lead
What's too risky
A human decidesFinance
Stripe · Wise
Watchdog
HubSpot · Slack
Head of Sales
What we say no to
A human decidesScreener
HubSpot
Scheduler
Calendar · Notion
A human owns the call An agent does the work, with only the tools that job needs
One question grades every task
What happens if it's wrong? 35 tasks in my company, three levels.
15
Fully autonomous
Wrong costs a minute. It runs, I read a report.
13
Human-assisted
Wrong is a bad draft. AI drafts, a person edits and sends.
7
Human-led
Wrong costs money or reputation. AI prepares, a human presses send.
Stop asking who to hire. Ask where a human really needs to decide.
10Stack
What I build with.
Chosen per problem rather than per preference. The list is short on purpose.
01Models & Agents
OpenAIAnthropicLangGraphLangChainCrewAILlamaIndex
02Retrieval & Memory
PineconeWeaviateChromaDBQdrantEmbeddingsEval harnesses
03Build
PythonFastAPITypeScriptReactNext.jsPostgreSQL
04Voice
VapiRetellTwilioElevenLabsDeepgram
05Product
JiraLinearMiroFigmaNotionDiscovery
06Infrastructure
AWSGCPSupabaseVercelDockern8n
11Evidence
Verified track record.
Quotes and ratings from completed client engagements — 101 of them, at a 100% Job Success score.
“Extremely high level of attention to detail, really obsessed about delivering high-quality results for the client, considerate and an expert in the field, as well as willing to get hands on.”
Detail OrientedCommitted to QualityProfessional
OpenAI prompt design & tool-use optimisation
Verified Upwork client · 5.0
“I highly recommend taking a consultation. Knows a lot about his industry.”
Clear CommunicatorAccountable for Outcomes
AI strategy consultation
Verified Upwork client · 5.0
Top 1% of talent on Upwork, screened by an independent industry expert.
Expert-Vetted
Invitation-only tierAcross 101 engagements in healthcare, legal, real estate and regulated enterprise.
$1M+Revenue delivered
What clients consistently flag
Endorsement tags selected on completed jobs
- Collaborative26×
- Clear Communicator19×
- Committed to Quality18×
- Reliable14×
- Professional10×
- Detail Oriented8×
12FAQ
Questions answered.
The things people actually ask before we get on a call.
13Next Step
Four ways to start.
Pick whichever fits where you are — whether you want it built for you or want to build it yourself. Each one ends with a straight answer rather than a proposal deck.
01
Book a Discovery Call
Thirty minutes on your problem, your users and your constraints. I'll tell you honestly whether I'm the right fit — I say no to roughly 40% of enquiries.
Schedule call30 min
02
Architecture & Feasibility Review
Already building? I'll review your AI architecture, retrieval quality and evaluation setup, and tell you where it will break before your users find out.
Request reviewFixed scope
03
Coaching for Founders
Want to build with AI yourself? One-to-one sessions on scoping, tooling, model choice and AI-assisted development, so you can ship without hiring an agency first.
Start coaching1:1 sessions
04
Send Me the Brief
Email your idea and rough budget. Within 24 hours you'll get a straight answer on fit, an outline scope, and two or three questions that matter.
Email meReply in 24h
12Contact
Let's talk.
Tell me what you're building and where it's stuck. I reply within 24 hours.