Healthcare
DRIS v1.0
Agentic pharmacovigilance platform. LangGraph agents on the Claude API, a SHA-256 signed audit log of every tool call, and human review of all clinical output.
In live use.
- LangGraph
- Claude API
- Signed audit log
- Human review
Engineering intelligence. Building impact.
REI I. ZAIDI
Agentic AI architect
- Enterprise Multi-Agent Autonomous Swarm Systems Agentic AI Architect
- Cognitive Runtime Synthesis & Ephemeral Micro-Harness Engineering
- MCP Transport Fabrics & Federated Decentralized A2A Economies Networks
- Interoperability Mesh Infrastructure
- Zero-Trust Micro-FinOps Tracing Attestation Governance & Multi-Tenant Cross-Silo Domain Security Platform
Senior AI Engineer at Infixio. Founder and CEO of Delta AI Engineering. Enterprise software since December 2010.
Engineering next-generation intelligent systems
Focus: Agents, Automation, Infrastructure, Governance, Real impact
- in software since December 2010
- 15 years 10 months
- production AI work began
- 2022
- NED University of Engineering and Technology, 2025
- MSc AI
- systems designed and catalogued
- 27
What REI stands for
Research, engineering and innovation, held to a single standard.
REI I. ZAIDI designs multi-agent and retrieval systems for production use in regulated and high-stakes settings. Every agent operates within an engineered harness that validates its output, scopes its tools, records each action and its cost, and requires human approval wherever the risk warrants it.
Research Engineering Intelligence
Rise Evolve Innovate
Research & Reasoning
Multi-agent systems, cognitive architectures, evaluations, model adaptation
Engineering & Enterprise
Agent harness engineering, MCP & A2A, LLMOps, AI infrastructure, secure and scalable systems
Innovation & Impact
Decentralized AI economies, interoperability, governance, real-world enterprise deployment
PeopleTechnologySolutionsImpact
- LangGraph
- Model Context Protocol
- A2A
- Claude API
- pgvector
- Hybrid retrieval
- Semantic reranking
- LangSmith
- Docker
- Kubernetes
- GitHub Actions
- FastAPI
- Neo4j
- LoRA / QLoRA
- PyTorch
- Power BI
Approach
Agentic systems that are verified before they act and accountable afterwards.
Eleven years of Microsoft enterprise delivery established a discipline of estimation, milestones and end-to-end lifecycle ownership. The same discipline governs every AI engagement: success measures are agreed first, one complete slice is delivered, and releases proceed only through evaluation gates, approval points and cost controls.
How a task moves through the harness
Interactive illustration · not live data
- Request
A task arrives with its context and the requester's permissions.
- Orchestrator
Produces a plan and assigns a specialist to each step.
- Specialists
Each specialist accesses only its authorised tools, through MCP.
RetrieveReasonAct - Evaluation gate
Output must pass schema validation and regression checks.
- Human approvalhuman in the loop
A designated reviewer signs off wherever the risk warrants it.
- Release
Released with a complete trace and cost record.
Trace
- ready: press Step or wait for the next task
Shows how a failed evaluation rolls the work back instead of shipping it.
Multi-agent systems
- ● In production
- Orchestrator, specialist and critic roles; reflection loops; human approval gates
- ● Built into products
- Routing that escalates instead of guessing; an RL controller coordinating five models (MARS-MINDS)
- ○ Studying now
- Deep Agents; CrewAI and AutoGen; ambient-agent patterns
Agent harness
- ● In production
- Validated structured output, regression evaluations as release gate, automated rollback
- ● Built into products
- Self-summarising session memory (TALEEM)
- ○ Studying now
- Temporal durable execution; gVisor sandboxes for tool runs
Tools and interoperability
- ● In production
- MCP servers and clients; provider routing with fallbacks
- ● Built into products
- Five-provider orchestration layer with SSE streaming
- ○ Studying now
- A2A protocol in production; agent payment protocols
Security and cost
- ● In production
- OAuth 2.0, JWT, RBAC, OWASP LLM Top 10 controls, per-tenant isolation, cost per successful task
- ● Built into products
- Signed tool-call audit log (DRIS); Merkle chain of custody (Sentinel)
- ○ Studying now
- Agent identities; OpenTelemetry; ISO/IEC 42001
Autonomy always operates within human approval gates. Items listed under "studying now" are not represented as delivered.
Verification protocol before release
Applied to every deliverable
The first 90 days of an agent project
- 1FrameDays 1–30
Map the workflows, data, tools and risks. Agree the success measures, the target cost per successful task and the points that require human approval.
- 2SliceDays 31–60
Deliver one complete vertical slice within a harness: end-to-end tracing, regression evaluations, validated output and scoped tool access.
- 3HardenDays 61–90
Add guardrails, the audit trail, tenant isolation, runbooks and handover. Conclude with a measured before-and-after report.
Products and platforms
Products and platforms across healthcare, education, security and enterprise.
The portfolio spans healthcare, special education, clinician wellbeing, tutoring, fraud intelligence, portfolio management and vehicle valuation. Each entry sets out what the system does and the technology behind it. Hover over a card, or select Preview, to see its interface; demos open in a new tab and run on sample data. Architecture and evaluation evidence is available on request.
Fraud intelligence
Sentinel
Investigation agents that correlate entities in a Neo4j graph, with HMAC-SHA256 honeytokens and an append-only Merkle chain of custody for evidence.
- LangGraph
- Neo4j
- FastAPI + SSE
- Merkle log
Portfolio management
Delta Horizon
Portfolio workspace for PMOs: portfolio and resource dashboards, an AI assistant panel, and Arabic and Urdu report views.
- PPM
- Dashboards
- Assistant panel
- Multilingual reports
Education
TALEEM
Multilingual tutoring with Wikipedia retrieval and streamed answers, plus adaptive session memory that summarises itself as a lesson grows.
- RAG
- Streaming
- Session memory
Education
LUMINA
Individualised education-plan drafting mapped to Pakistani and international school boards, with streamed teaching responses.
- Streaming
- IEP drafting
- Curriculum mapping
Healthcare & special education
Zeta-9
Platform for a credential-verified network of healthcare and special-education professionals: screening support, IEP drafting and teleconsultation.
AI as decision support only, never diagnosis. Selected as one of 24 start-ups for ASPIRE Pakistan NIB IV pre-incubation.
- Screening support
- IEP drafting
- Teleconsultation
Healthcare & special education
Minds platform
The parent and professional front door for Zeta-9: one developmental profile shared by therapists, special educators and families, with screening support and IEP drafting.
Screening support only, never diagnosis.
- Multilingual UI
- Shared profiles
- Role-based views
Clinician wellbeing
RAHAT-AI
Voice-first burnout early-warning system for hospital teams, usable in Urdu, Punjabi, Sindhi, Pashto, Balochi, Saraiki and English, with confidential referral rather than supervisor alerts.
Built for the Rufayda Anti-Burnout Hackathon 2026.
- Voice interface
- Multilingual
- Ward risk view
Automotive
VehicleIQ
Plain-language vehicle valuation: describe a car in English or Arabic and get a price range from comparable vehicles, with a downloadable PDF quotation.
- Agentic valuation
- English + Arabic
- PDF quotation
Healthcare analytics
Autism screening analytics
Clinician-readable Power BI reporting built from 4,460 autism-screening records from Pakistan (2024).
Screening analytics, not diagnosis.
- Power BI
- Data modelling
Planetary AI
MARS-MINDS
MSc thesis: deep learning on public NASA HiRISE orbital imagery for Martian terrain classification, dust-storm analysis and mission-planning support (MarsHybridNet).
Six manuscripts in preparation; none presented as published. Public data; no space-agency affiliation.
- PyTorch
- EfficientNetV2
- ConvNeXtV2
- HiRISE
Career
From enterprise .NET delivery to production agentic AI.
15 years 10 months in software: enterprise engineering at Techerz, ExaTex and Learning Pitch, followed by applied AI at Infixio and at Delta AI Engineering, which she founded in 2022.
- Techerz
- ExaTex
- Learning Pitch
- Delta AI Engineering
- Infixio
- AI residency
- Enterprise engineering
- Applied AI
- Founder
- Teaching
Sep 2026 – present
Co-Team Lead
AI Internship Programme (Batch 1), with NCBC and HPCC, NED University
- A 12-week hybrid residency delivered under the AQLENTIS programme name.
- Co-designed PADDLE, a six-stage agent-design method: Problem, Action space, Data, Decision logic, Loop, Evaluation.
Nov 2023 – present
Senior AI Engineer · Applied Data Scientist
Infixio
- Builds production multi-agent platforms on LangGraph with orchestrator, specialist and critic roles and human approval gates.
- Uses regression evaluations as the release gate, with automated rollback and LangSmith tracing of every run.
- Integrates tools through Model Context Protocol servers; routes across Claude, GPT, Gemini and Mistral against cost and latency budgets.
Jan 2022 – present
Founder & CEO
Delta AI Engineering
- Sets strategy, architecture and governance for healthcare, education, fraud-intelligence and portfolio products.
Feb 2019 – Oct 2023
Software Architect · Lead Software Engineer
Learning Pitch
- Owned the full SDLC across concurrent client engagements and digital-transformation programmes.
- Led and mentored engineering teams; ran pre-sales with system integrators, wrote SOPs and ran implementation audits.
Aug 2015 – Jan 2019
Senior Software Engineer
ExaTex Enterprise Solutions
- Customised Microsoft Project Server 2010/2013 and SharePoint solutions, workflows, portals and SSRS reporting.
- Ran requirement workshops, UAT, deployment and end-user training; wrote SRS and architecture documents.
Dec 2010 – Jul 2015
Software Engineer · Junior Software Engineer
Techerz IT Consultancy Firm
- Led .NET web-application teams: planning, effort estimation, milestones, data modelling and integrations.
- Wrote proposals, specifications and runbooks; ran technical interviews and in-house training.
Research
MARS-MINDS: interpreting Martian terrain from orbit.
Her MSc thesis, the Martian Intelligent Navigation and Decision System, applies deep learning to public NASA HiRISE orbital imagery to classify terrain, support dust-storm analysis and inform mission planning under uncertainty.
- Models: MarsHybridNet, with EfficientNetV2, ConvNeXtV2 and InceptionV3 baselines
- Questions: terrain understanding, dust storms, habitat suitability, landing-site assessment
- Status: six manuscripts in preparation; none presented as published
HiRISE images are public data. This work has no affiliation with NASA or any space agency.
Credentials
Education, training and recognition.
Education
MSc Artificial Intelligence
NED University of Engineering and Technology, Karachi
2023–2025
BS Computer Science
Hamdard University, Karachi
2007–2010
Degrees and experience letters were issued under the name Ruksh E Ibadat.
Professional development
- Certificate of Completion: Free Introduction to Cognitive Project Management in AI (CPMAI)™, Project Management Institute, Oct 2026. A course completion, not the PMI-CPMAI™ certification.
- GenAI to Agentic AI: Full-Stack Development, NCBC, HPCC and NED University, Feb–Jun 2026
- ASPIRE Silicon Valley Cofounder Training, Batch 2, Aug 2026
- AI Agents Core Components, O'Reilly, Jun 2026
Health and education training
Behavioural science and child development inform how she designs health and education tools: AI supports screening and professional judgement, and never replaces diagnosis.
- Early Childhood Education Diploma, IECED / SBTE Karachi, 2017–18
- CBT Certificate Diploma (2025) and DBT skills course (2026), MedicPro International Academy, UK
- RBT® 40-hour training (2022), which is training and not RBT certification
- 2025 workshops: Child PsyCon (KEMU / PPRC), ADHD awareness, neurodivergent behaviour
Training and participation only; REI holds no clinical licence.
Recognition and community
- Finalist, ASPIRE Pakistan National Idea Bank IV pre-incubation, with Zeta-9: one of 24 start-ups selected.
- Built a workshop programme on AI in health for ICCBS and SIUT, Karachi.
The ASI Society
Practical skills. Real use cases. Measured claims.
A private community led by REI for engineers, researchers, founders and learners applying AI in their work, products and ventures. Members share findings, credit their sources and report what did not work.
Membership is by request. Please send a brief introduction and what you intend to learn, build or contribute.
Contact
Discuss an engagement.
REI welcomes enquiries about agentic AI architecture and AI platform leadership roles, partnerships on agent harnesses, evaluation and governance, and research collaboration. Available on-site with relocation, hybrid or remote.
- Based in
- Karachi, Pakistan · UTC+5 · English, Urdu