AI Product and Automation Engineer. Based in Islamabad, Pakistan.
AI SaaS products, production LLM & RAG systems, workflow automation & agent orchestration for founders and operations teams
AI Product & Automation Engineer
Automated manual workflows at a US wealth firm, shipped a healthtech product in the UK, automated retail hardware in Tokyo—all in use today.
Location: Islamabad, PAK
AI Product & Automation Engineer
AI SaaS products, production LLM & RAG systems, workflow automation & agent orchestration for founders and operations teams
Location: Islamabad, PAK
AI PRODUCT & AUTOMATION ENGINEER Ammad Ashraf — shipping software since 2024, now engineering AI products, automations and the agentic systems around them, live for 150+ users across four countries.
Let's build the next one.
Expertise
Tools
Model training, retrieval pipelines, and the orchestration layers that carry LLMs into production.
AI & agents
Work
01
PrediabetesBasket — receipt analysis pipeline
AI Automation
- Health guidance
- Consumer SaaS
- Email-driven pipeline
- United Kingdom
Customers forward a supermarket receipt and get a full basket breakdown back before they have put the shopping away — and when something goes wrong, they are told exactly what and why.
02
An AI prediabetes app that scores your weekly shop
AI Product Engineering
- Health guidance
- Consumer SaaS
- Mobile-first web
- United Kingdom
For UK adults with prediabetes. Forward a supermarket receipt: AI rates every item, scores the basket, suggests swaps and tracks progress. Free once, then monthly or yearly.
03
Heritage AI Platform — the document-processing engine
AI Automation
- Wealth management / legaltech
- Document-intelligence pipeline
- n8n + LLM backend
- United States
A wealth-management firm’s legal documents arrive as PDFs of wildly different shapes — wills, trusts, insurance policies, powers of attorney. This is the engine that reads them, classifies them, and writes them into a database an operations team can trust.
04
Heritage AI Platform — the operations dashboard
AI Product Engineering
- Wealth management / legaltech
- Internal SaaS platform
- Web dashboard
- United States
The wealth-management operations team that used to retype legal documents into Word and PowerPoint by hand now works inside one dashboard — reviewing what the AI extracted, editing it inline, and exporting a finished book with one click.
05
ContraGuard AI — 5-agent legal-contract risk analysis
Machine Learning
- Legal risk analysis
- Academic / personal project
- Web app (FastAPI + React)
- Pakistan
Upload a PDF contract and five agents work through it in sequence: parse, classify every clause into one of 41 legal categories, score its risk against a reference corpus, explain the flagged ones, and generate a colour-coded PDF report — validated against a real freelance agreement, not a synthetic test.
06
Capital Transit — an agentic trip planner for Islamabad’s bus network
Agentic Systems
- Public transit / urban mobility
- Academic / personal project
- Web app (FastAPI + React, SSE)
- Pakistan
Ask it how to get across Islamabad by CDA bus, in plain language, misspelled stop names and all — the assistant calls the right tools in the right order, corrects your spelling, and streams back a real route with real departure times.
Journey
- Sep 2025PresentEmployment
AI Solutions Engineer (Full-Stack & Automation)
Shipped two production AI products — a document platform 100+ people use, cutting processing time by around 80%, and a UK health-tech SaaS that reached 50+ users and 5+ paid subscribers in its first two weeks.Ownership- Heritage AI Platform — owned the full lifecycle of a document-intelligence platform for a US wealth-management firm.
- Heritage AI Platform — built the n8n ingestion pipeline: 8+ legal document types, one dispatch workflow, 4 modules, 30+ parallel extraction pipelines.
- Heritage AI Platform — engineered the PostgreSQL model (65+ tables) with entity resolution across 15+ entity types and confidence-scored, provenance-tracked JSONB.
- Heritage AI Platform — hardened it for production: full RLS, JWT-secured Edge Functions, 88 automated tests, query optimization.
- PrediabetesBasket — owned the full product lifecycle of a UK health-tech SaaS that scans grocery receipts against dietary guidelines.
- PrediabetesBasket — built the n8n pipeline: 8-flag medical safety gate, LLM analysis, 4 FK-dependent writes, zero silent failure paths.
- PrediabetesBasket — built Stripe billing (£7.99/mo, £59/yr) on 5 Supabase Edge Functions, plus 7 transactional email templates.
Collaboration- Sole engineer on the Heritage AI Platform for seven months, syncing weekly with the CEO in Oslo.
- Architecture calls made directly with the wealth-management operations team who use it daily.
- RBAC validation screens shipped with the ops leads so each role edits only its own data.
- PrediabetesBasket — delivered solo, with a product owner and delivery manager on scope.
Milestones- Heritage AI Platform in production for 100+ active users — around 80% less time per document.
- Manual Word/PowerPoint workflow retired; Heritage Books export as a one-click PDF.
- Security 2/10 → 8/10, reliability 3/10 → 7/10; queries per page load 23–35 → 5–8.
- PrediabetesBasket live — 50+ users and 5+ paid subscribers within two weeks of launch.
- Large-receipt LLM latency down 67% (4m30s → 1m30s); standard receipts down 25% (60s → 45s).
- Feb 2026PresentResearch
Co-Author, DeepTrust Deepfake Detection
Co-authored a multimodal audio-visual deepfake-detection paper, submitted to IBCAST 2026 — and is now building the research into a product, SyncLens, with a first commercial launch targeted for September–October 2026.Ownership- Co-designed SyncCAM, an 8-head bidirectional cross-attention module fusing a 3,328-D visual embedding with a 2,816-D audio embedding.
- Kept every pretrained backbone frozen — ablation-confirmed — for calibrated 4-class classification on the ~21.5K-video FakeAVCeleb v1.2 benchmark.
- Individually built the Statistical Attack dataset generator — JPEG compression, Gaussian noise, DCT manipulation, colour-histogram shift — to stress-test the detector.
- Designed the white-box adversarial track (FGSM/PGD on a ResNet surrogate) and led the cross-dataset generalization study across AV-Deepfake1M and DFDC.
Collaboration- Co-authored with Palwasha Zahid, Dr. Qurat Ul Ain (supervisor) and Noor Ul Baseer.
- A registered FAST-NUCES final-year project, supervised by Dr. Qurat Ul Ain.
- Building SyncLens independently — API and GPU-backed inference — for remote hiring, trust & safety, forensics, journalism and developer integrations.
Milestones- Paper submitted to the 23rd IBCAST 2026 — under revision ahead of resubmission with the adversarial and generalization results.
- Cross-dataset study surfaced a critical generalization gap in the base model.
- SyncLens first commercial launch targeted for September–October 2026.
- Jul 2026Freelance
AI Automation Engineer
Shipped a production n8n SEO-analysis workflow inside a paid 72-hour technical assessment, self-taught from sparse documentation, delivered with a full technical report and live Google Sheet output.Ownership- Built the analyze_seo_page n8n skill — HTML fetch, 8,000-character truncation, and a constrained OpenAI prompt returning a stable 5-field JSON contract.
- Structured it as 9 sequential steps mirroring the skill's own instruction protocol, from HTML fetch to a validated result.
- Designed a typed error-classification system feeding one downstream error-handler node.
- Built defensive output normalisation so every downstream node gets the same 8-key shape regardless of model output quality.
Collaboration- Solo submission against a fixed 72-hour paid assessment window.
- Learned the "OpenClaw" skill-authoring convention from sparse documentation with no prior exposure.
Milestones- Delivered inside the 72-hour deadline: working code, technical report, walkthrough, live Google Sheet and an agent-skill markdown file.
- Results land in the Google Sheet with zero post-processing.
Links - Jun 2026Freelance
Agentic Software Engineer
Delivered a proof-of-concept AI Concierge backend for a US travel-loyalty platform — an MCP-exposed recommendations API with a fully tested, auditable rule engine — in a scored, client-evaluated freelance engagement.Ownership- Architected a multi-tenant travel-recommendations API exposed over the Model Context Protocol via HTTP/SSE, alongside a parallel REST surface.
- Enforced tenant isolation across 3 partner configurations, rejecting any partner/member ID mismatch with an explicit PartnerContextMismatchError.
- Built a deterministic RuleEnforcer isolating partner business rules completely from the transport layer, with a documented exclusion-before-cap ordering.
- Wrote 28 isolated Vitest unit tests exercising the rule engine and scoring logic independently of any transport.
Collaboration- Sole engineer on a scoped, evaluated 4-week freelance technical assessment.
- Documented and justified every AI-assisted design decision for client review, including one caught AI-drafted bug.
Milestones- 28/28 unit tests passing, including an adversarial "excluded item still scores highest" case.
- Tenant isolation held across 3 partner configurations.
- Fully containerised proof-of-concept with a live MCP tool-call trace demo.
Links - Mar 2026Apr 2026Employment
AI Automation Engineer
Replaced a manual Excel reporting process with nightly automated runs across five report types and several client workspaces, with no data loss and a new client onboarded by config alone.Ownership- Built the Node.js ETL pipeline connecting SmartAdvocate (SSRS) to Zoho Analytics on email-triggered ingestion.
- Designed the fault-tolerant architecture: structured logging, duplicate prevention by processed-email state, per-report import strategy, failure alerting.
- Wrote the schema-driven config engine — one generic processor covering 5+ report types with no report-specific code.
Collaboration- Worked across several client workspaces with the consulting team defining each report’s shape.
- Handed over config-only onboarding so a non-engineer can add the next client.
Milestones- Manual Excel reporting replaced by nightly automated runs.
- Zero data loss across nightly pipeline runs.
- New client onboarded through a config change alone.
- Oct 2025Dec 2025Internship
Software Engineer (Full-Stack)
Made an unmanned shop work end to end: a customer’s payment reaches a dispenser in a physical store with no human step in between.Ownership- Built the backend for an unmanned retail environment — order processing and payments on the Square Node.js SDK.
- Implemented secure webhook processing with signature verification against Square Cloud APIs.
- Held full-duplex WebSocket connections from the cloud backend to in-store local servers driving hardware dispensers.
Collaboration- 46 days onsite in Tokyo, with formal training in intercultural collaboration, compliance and Japanese business culture.
- Worked alongside peers from Asia, Africa, Latin America and Europe, and with hardware engineers to bridge cloud events to physical dispensers.
Milestones- Selected as 1 of ~50 engineers worldwide, from roughly 14,000 applicants, for METI Japan’s AI and Tech Talent internship.
- Payment reaches a dispenser with no human step in between.
- 100% data integrity across Visa and Apple Pay checkout flows.
- Remote hardware triggering live across multiple physical stores.
Links - Aug 2025Oct 2025Internship
Solutions Engineer
Built and shipped the developer portals behind enterprise proofs of concept for JPMorgan, Shell, Vodafone and Visa.Ownership- Converted client API specifications into functional developer-experience portals with the APIMatic CLI.
- Ran SDK compilation and end-to-end testing across Java, Python, TypeScript, Go and C#.
- Automated CI/CD with GitHub Actions and deployed portals to production via Cloudflare.
Collaboration- Embedded in the Solutions Engineering team on enterprise proofs of concept.
- Authored API recipes and workflow documentation used in technical sales conversations.
Milestones- Portals shipped behind PoCs for JPMorgan, Shell, Vodafone and Visa.
- API Copilot and Cursor integrations live in delivered portals.
Links - Jun 2025Sep 2025Venture
Co-Founder
Co-founded an AI idea-validation platform for early-stage founders and designed its evaluation matrix from structured interviews at the National Incubation Center in Islamabad.Ownership- Co-founded Clarimo AI, an AI workspace that consolidates idea validation, founder pain-point research and evaluation scoring into one single platform — helping new founders systematically validate their ideas before they spend months building something nobody wants.
- Designed a founder-facing idea-evaluation matrix, built from structured interviews with the National Incubation Center in Islamabad, Pakistan.
Collaboration- A student-led initiative, co-founded with two teammates.
Milestones- Evaluation matrix built from structured interviews with the National Incubation Center in Islamabad.
- Stepped away in late Sep 2025 for the onsite placement in Tokyo.
- Apr 2025Aug 2025Venture
Co-Founder
Co-founded a software agency delivering business automation — connecting the tools clients already run so the manual step between them disappears.Ownership- Co-founded SingleCore, a services-based automation agency providing business-automation solutions and custom development services.
- Oversaw scoping, delivery and client handoff end to end across every engagement.
Collaboration- Founded with three university colleagues.
Milestones- Agency operating from April to August 2025.
- Manual steps between clients’ existing tools removed.
- Jan 2025Feb 2025Internship
Backend Engineer
Built the internal system 10+ people used to onboard, track and review interns, with deadline reminders that chase people so a manager does not have to — cutting manual workload by around 60%.Ownership- Engineered the Intern Management System — onboarding, task tracking and performance evaluation.
- Implemented JWT-based RBAC across two admin levels with unauthorised-action prevention.
- Built real-time chat and notifications over WebSockets with heartbeat pings and room-based logic.
- Added Node-cron schedulers for deadline monitoring and task-specific push notifications.
Collaboration- Built against the working requirements of the 10+ staff who run the intern programme.
- Handed over a modular Express architecture with validators, rate limiting and auth middleware.
Milestones- 10+ active users on the internal system.
- Manual workload cut by around 60% — deadline reminders chase people so a manager does not have to.
Links - Aug 2024Dec 2024Internship
Backend Engineer
Owned the certificate workflow in a learning platform end to end — issuing, editing, approval and download, with each role able to touch only its own part.Ownership- Led the certificate workflow module — creation, editing, draft saving and approval.
- Built dynamic certificate customisation: managers edit predefined fields, layouts stay fixed.
- Shipped admin approval and download with RBAC, AES encryption and JWT auth.
Collaboration- Contributed to a role-based LMS spanning admins, managers, students and mentors.
- Designed the core class diagram with the team using GoF patterns across Controller, Service and DAO layers.
Milestones- Certificate workflow owned end to end.
- Each role able to touch only its own part.
Links
Services
Replacing one manual process end to end so it runs itself.
Operations teams, agencies and small companies with a process that repeats every week.
- Process Audit$2503 business days
- Best for
- You know something is slow but not what to automate first.
Includes- One process mapped step by step
- The hours and cost of today’s process, measured
- What to automate and what to leave alone
- A written plan with a fixed quote
OutcomeYou know what to build, what it costs and what it saves before you commit.
NoteCredited to the build if you go ahead within 30 days.
- RecommendedOne Process, Automatedfrom$1,0001–3 weeks
- Best for
- One clear process, one clear owner.
Includes- One process automated end to end, across up to 3 tools
- Error handling and failure alerts
- A run log, so you can see what happened and when
- Video walkthrough and written handoff
- 14 days of fixes included
OutcomeOne process that no longer needs a person.
- Automation Systemfrom$3,0004–8 weeks
- Best for
- Several connected processes, or one that has to scale to many clients.
Includes- 3–5 connected workflows
- Built to be configured, not rewritten: new cases without new code
- Monitoring and a simple status view
- Team training session
- 30 days of fixes included
OutcomeA small automation layer your team can run and extend.
From idea to paying users, built to hold up past the demo.
Founders with an idea, or a fragile prototype, and no engineering team.
- Production Rescuefrom$1,0001–2 weeks
- Best for
- A prototype built with Lovable, Bolt, Cursor or similar that now needs to be safe for real users.
Includes- Security review: logins, permissions, who can see what
- Fixes for the issues found, up to the agreed scope
- Stripe billing wired up and tested
- Error tracking and basic automated tests
- A written report of what changed and why
OutcomeA prototype that survives real users.
NoteFixed price is confirmed after a short review of your code.
- RecommendedLaunch-Ready MVPfrom$3,5005–8 weeks
- Best for
- A clear idea with one main user journey.
Includes- The core product: one main journey, up to about 8 screens
- Accounts with two roles: user and admin
- Stripe subscriptions on up to two plans
- One AI feature with safety limits and cost caps
- Basic analytics and transactional emails
- Deployed live, with docs and handoff
OutcomeA product you can put in front of paying customers.
- Product Partnerfrom$6,00010–14 weeks
- Best for
- A bigger product, or one that has to plug into other systems.
Includes- Everything in the MVP, plus up to 3 integrations
- Team accounts and multi-role access
- AI pipeline with monitoring and quality checks
- Weekly demos and a shared roadmap
- Optional monthly build retainer after launch
OutcomeA full product and an engineer who stays accountable for it.
AI that finishes multi-step jobs, and asks a human when unsure.
Teams whose AI works in a demo but has not yet been trusted with a whole workflow.
- Agent Feasibility Sprint$7501 week
- Best for
- You are not sure AI can do this job yet.
Includes- One workflow picked and scoped
- A working prototype on your sample data
- Accuracy tested on 30–50 of your real cases
- A build or don’t-build recommendation with a cost estimate
OutcomeAn honest answer to “can AI do this?” before you spend real money.
NoteCredited to the build if you go ahead within 30 days.
- RecommendedProduction Agent Workflowfrom$2,5003–5 weeks
- Best for
- One multi-step workflow that is ready to go live.
Includes- One agent workflow, live: up to 5 steps and 3 tools
- Human-approval steps where stakes are high
- Retries, logging and cost limits
- A test set that catches regressions
- A runbook: how to monitor it and fix it
OutcomeA workflow you can trust to run without watching it.
- Multi-Agent Systemfrom$5,0006–10 weeks
- Best for
- Several coordinated agents across a larger process.
Includes- Up to 3 coordinated agents with routing and shared state
- A review dashboard for people in the loop
- A trace viewer: see every step of every run
- Load and failure testing
- Team handoff and training
OutcomeA whole process run by agents, with people supervising.
Turning your documents into answers you can check.
Firms buried in documents: legal, finance, health and operations teams.
- Document Intelligence Pilotfrom$1,5002–3 weeks
- Best for
- One document type you handle by hand today.
Includes- One document type turned into structured fields
- A confidence score on every field
- Tested on up to 50 of your real samples, with an accuracy report
- A simple review screen or spreadsheet export
- Clear notes on what changes at scale
OutcomeProof, on your own documents, that the system works.
- RecommendedKnowledge Assistantfrom$2,5003–5 weeks
- Best for
- A team that keeps asking the same questions of the same documents.
Includes- Answers cited to the exact document and section
- Up to about 200 documents, agreed upfront
- Access rules: people only see what they are allowed to
- A test set of real questions to measure accuracy
- Deployed, with cost and quality monitoring
OutcomeAsk a question in plain English, get an answer with its source.
- Production Document Platformfrom$6,0008–12 weeks
- Best for
- Many document types, many users, and a real need for audit trails.
Includes- Up to 4 document types, routed automatically
- A review dashboard with role-based access
- Records kept consistent across documents (entity matching)
- An audit trail of who changed what
- Security hardening and automated tests
- Documentation and handoff
OutcomeA document workflow the whole team uses every day.
LLM speed and cost tune-up: from $750, one week. Find out why an AI feature is slow or expensive, then fix it.
Letting AI assistants use your product, within your rules.
Product companies and platforms that want AI assistants to use their service safely.
- MCP Readiness Review$3003 business days
- Best for
- You are not sure whether to build one, or what it should be allowed to do.
Includes- An inventory of the tools and data worth exposing
- A security review: sign-in, permissions, prompt-injection risks
- The rules that must never be left to the AI
- A build plan with a fixed quote
OutcomeA clear yes or no, and a safe plan if it is a yes.
NoteCredited to the build if you go ahead within 30 days.
- RecommendedMCP Serverfrom$1,5002–3 weeks
- Best for
- One product or API you want assistants to be able to use.
Includes- One MCP server exposing 3–5 tools from your API or database
- Strict input checks and clear error messages
- Sign-in and permissions set up
- A test suite and a setup guide
- A live demo in Claude with a trace of every tool call
OutcomeYour product, usable from an AI assistant.
- Multi-Tenant MCP and Rules Enginefrom$3,5004–6 weeks
- Best for
- Several partners or customers, each with their own rules.
Includes- REST API and MCP server sharing one set of business logic
- A deterministic rule engine, tested on its own
- Tenant isolation for up to 3 partner configurations
- An audit trace on every response
- A caching layer and admin controls
- Dockerised, with docs and handoff
OutcomeAI assistants that follow each partner’s rules every time.
None of these quite fit? Describe the problem and I will tell you honestly whether I am the right person for it.
Start a conversationBeyond a fixed project
- Hourly · ad hoc$25/hr5-hour minimum
Small jobs, code reviews, rescue work and quick advice.
- Care retainer$400/mo16 hours a month
Keeping something I built healthy.
- Monitoring and fixes
- Small changes and tweaks
- Reply within 1 business day
- Unused hours roll over once
- RecommendedBuild retainer$900/mo40 hours a month (about 10 a week)
Steady build work on one project.
- One workstream, shipped every week
- Weekly call and written update
- Reply within 1 business day
- Unused hours roll over once
- Embedded retainer$1,500/mo80 hours a month (up to 20 a week)
An engineer inside your team.
- Up to two workstreams
- Daily written updates and a weekly planning call
- Same-day replies during overlap hours
- Unused hours roll over once
Terms & how billing works
- Phase 01
Before we start
PricingFixed priceIf new requirements come up along the way, I’ll scope and quote the addition — billed at $25/hr.
AvailabilityUS · UK · EUI’m happy to sign an NDA before we begin.
- Phase 02
While I build
Payment scheduleIn milestonesUp to $2,00050 start · 50 deliveryLarger projects40 start · 30 midpoint · 30 delivery - Phase 03
On delivery
Ownership100% yoursYou own the code, prompts, workflows and documentation once the final payment lands.
No lock-inYour stackWherever possible, I build directly in your own accounts and repositories.
- Phase 04
After launch
Warranty14–30 daysEvery project includes a support window after delivery — 14 days on single & 30 days on larger builds.
Cancellation30 days’ noticeRetainers can be paused or cancelled anytime with 30 days’ notice.
Impact
Endorsement
01
Ammad has done great work for us, helping our agency build and deliver complex client applications and automations. He consistently writes clean, structured, and maintainable code.
Ammad has a strong technical grasp of modern development stacks and handles everything from architecture to final polish with great care. Beyond his technical skills, he is reliable, communicates proactively, and ensures products are fully ready for handoff and real-world use.
Any team or agency looking for a talented developer who can hit the ground running would be lucky to have Ammad on board!
Endorsement
02
Ammad joined our Tokyo team as one of two interns selected for our project under METI’s AI and Tech Talent Internship Program (Ministry of Economy, Trade and Industry, Government of Japan). I supervised him directly, and we quickly built a great working relationship. He was very friendly, communicated very well with us, and formed strong connections with the entire team.
On the technical side, he took a very practical and respectful approach to our codebase. He made sure to properly understand our legacy systems first before independently integrating the new systems into them.
We truly appreciated his adaptability and the positive presence he brought to our office.
Endorsement
03
Having known Ammad closely over the last three years, I’ve had the chance to see firsthand how he thinks about technology.
He approaches engineering with a very clear product and entrepreneurial mindset. Usually, our discussions start with him identifying a specific problem and quickly turn into him designing the architecture to solve it.
He is incredibly curious and constantly bringing new ideas to the table. It’s rare to find someone who can so naturally connect a raw business concept directly to a practical, working solution.
Endorsement
04
I managed him directly during his APIMatic internship, and his ability to learn unfamiliar systems quickly was exactly what we needed.
We tasked him with building API proofs of concept and developer portals for major enterprise clients. He quickly developed a strong grasp of our entire product ecosystem and understood how all the moving parts fit together.
He is the kind of engineer who can jump into a complex technical environment, figure out the architecture rapidly, and just get straight to work.
Endorsement
05
In his final-year research, Ammad showed a strong ability to connect academic concepts directly with product engineering. He took the lead on the AI development for the group, handling the model training and building the actual data pipelines.
He collaborated really well with the rest of the team. His research approach was highly practical—he consistently applied solid AI engineering principles to ensure our work was structured for real-world application rather than just staying in the lab.
Engage
Know exactly what you need? Pick a tier below. Still shaping the idea? Skip this and just tell me what’s on your mind.
Let's make your business AI native.