NowData & AI Engineer, Saizen AcuityBEng & MEng, Heriot-Watt
SankithaKirindu
AI systemsData infrastructureAutomation platformsFull-stack product
Systems that keep working after launch.
I build the production engines that run Saizen Acuity's own operation, and the platforms it delivers for clients in property, recruitment and hardware. First in the country for Cambridge A-Level Computer Science, now on Heriot-Watt's highest merit scholarship.
Before I build anything, I want to know why the problem exists.
I design and deliver production AI systems, data infrastructure and commercial automation platforms for UK and international businesses. At Saizen Acuity, a London AI, data and automation consultancy, that means the engines behind its own operation and the platforms it builds for clients in property, recruitment and hardware.
The first thing I ever fixed was my grandmother's Nokia, the day before I turned three. The habit stuck. Then one of my workflows failed silently in production, and a senior engineer asked what happens when an assumption is wrong. I have asked it of every system since.
The question he asks of every systemWhat happens when an assumption is wrong?
Rule 01
Validation before confidence
Inputs are checked against what the system expects before any model sees them. Bad data stops at the door, loudly.
Rule 02
Fallbacks before elegance
Timeouts retry, malformed outputs get repaired, and a second path exists before the first one is made beautiful.
Rule 03
Human review before anything irreversible
Anything that cannot be taken back, a client email or a public post, waits for a person to approve it.
A web development internship in July 2025 became full-time responsibility for AI automation, data engineering, analytics and client delivery. Select any piece of work on the timeline for the full story: situation, task, action, result.
Jul 2025OctJan 2026AprJulOct
RolesSaizen Acuity, remote
InternData & AI Engineer, full-time
Now
+270%page views against the old baseline
Jul 2025Web Developer & Programmer Intern
The website relaunch
Situation
The company website was a static presence, not a measurable business system.
Task
Rebuild and relaunch it: design, development, SEO, analytics and security.
Action
Rebuilt the pages with custom CSS and JavaScript, structured them around clear journeys, implemented schema and on-page SEO, set up GA4, heatmaps and event tracking, and hardened SSL, access, firewall and backups.
Result
+270% page views, +119% sessions, +69% visitors, +46% average session duration, and bounce rate down 27%.
300+articles a day through the news system
Aug 2025Web Developer & Programmer Intern
From brochure to system
Situation
A relaunched site still has to do work after the visitor arrives.
Task
Turn the website into the foundation the company's later systems could build on.
Action
Built reusable landing-page frameworks, member and AI Readiness Review pages, a news system, and form routing into Slack and Google Sheets through n8n.
Result
A news system pulling 300+ articles a day, and the base the automation engines were later built on. A month in, the internship became a full-time role.
15+engines in production on one standard
2025 – nowData & AI Engineer
The automation engines
Situation
Research, content, follow-up and delivery ran on manual effort, and every new workflow risked becoming its own little universe.
Task
Build engines that keep working every day after launch, on one shared standard.
Action
Built 15+ engines across research, lead generation, content, email and client delivery on one standard for ingestion, reasoning, structured output, monitoring, retries and recovery. Designed the 14-assistant Saizen GPT Suite alongside it, each with a fixed output contract.
Result
About 1,200 tasks a day, 3,600+ reports a month and an estimated 5,000+ hours saved a month. Deploying a new client engine went from hours to minutes.
50 / 50tracked keywords improved in 45 days
Dec 2025 – Jul 2026Client: Prestige Properties Cyprus
Prestige Properties Cyprus
Situation
A property website stops being "just design" very quickly: design, listings, CRM and data had to become one platform.
Task
Lead the technical rebuild end to end, then make the new site findable.
Action
Led the frontend, listing architecture, enquiry journeys and launch. Reconciled 457 property records for the website, Property Hive and Rightmove, wired every enquiry into a CRM running 13 automation workflows, then ran a 45-day SEO programme.
Result
All 50 tracked keywords improved: average rank 44.4 → 9.4, Google Top 10 from 1 keyword to 40, and 9th to 2nd against eight competitors.
~20,000qualified leads from 100,000+ raw records
Jan – Jun 2026Data engineering and GTM
Lead qualification and Vectis.ai
Situation
Lead generation fails when raw volume is mistaken for value.
Task
Find the records worth acting on: first in 100,000+ raw businesses, then in one concentrated market, a hardware trade event.
Action
Designed and ran the pipeline: extraction, normalisation, deduplication and verification, then wrote the rules for what counts as commercially usable. For Vectis.ai at Hardware Pioneers MAX, researched 253 exhibitors and mapped ~700 profiles by strategic fit.
Result
~20,000 qualified leads in 16 sectors, ready for CRM import, and 35 priority accounts with prepared intelligence, including Jaguar Land Rover, STMicroelectronics and Medtronic Digital Surgery.
A recruitment brand, a client-delivery process scattered across email and shared drives, and a company website that had to behave like part of the company.
Task
Ship each one, and keep it coherent while the brief kept moving.
Action
Built ByteHire's complete frontend through ~118 tracked revisions. Designed the access model and built the backend, database and messaging of a Flask customer portal. Engineered Saizen Acuity V2 as one Flask application.
Result
A customer portal live with 3 access levels and 0–100% project tracking, and Saizen Acuity V2 live at saizenacuity.com from November 2026.
Fifteen-plus engines run research, content, follow-up and delivery for Saizen Acuity and its clients. Pick one, break something, and watch the standard catch it. Then switch the standard off and see what a silent failure looks like.
15+engines in production
1,200+tasks a day
3,600+reports a month
Minutesto deploy a new client engine, down from hours
1. Pick an engine
2. Break something
Person Research Engine
Trigger
Gather
Validate
Reason
Shape
Review
Deliver
Monitor
Held for a personClient email ready to send
What happened
0runs
0faults caught
0retries
0held for a person
0silent failures
An illustrative run of the standard, written from how the engines are built. It is a simulation, not live data from a client system.
05Selected projects
Ten of thirty-six, strongest first.
Each opens to its full case study on sankitha.com. Three of them run here: find an equation, qualify a hundred thousand records, and move a component without breaking its connections.
20 adversarial functions (oscillation, discontinuities, nested forms) against an engine that searches polynomial, rational, exponential, trigonometric and composite forms for a readable equation.
9 exact structures plus 1 equivalent recovered on the retest, over 230 training and 86 holdout points; the write-up lists the failures too.
A curve fit tells you what the data did. An equation is an argument about why, and it can be checked.
QuestionRetrieve from 37,613 entriesCalculate in codeModel explains
Accuracy on 200 hand-graded questions over a frozen record.
03
Prestige Properties Cyprus
Real-estate platform: rebuild, data, CRM and search
Dec 2025 – Jul 2026Custom CSS & JavaScript · Property data · CRM automation · Technical SEO
457 property records cleaned and every enquiry wired into a CRM running 13 automation workflows, after leading the technical rebuild: frontend, listing architecture, enquiry journeys and launch.
50 of 50 tracked keywords improved in a 45-day SEO programme: average rank 44.4 → 9.4, and 9th to 2nd against eight competitors.
A property website stops being "just design" very quickly: design, listings, CRM and data had to become one platform.
Only 100,000+ in and ~20,000 out are from the project. The stage counts and the effect of skipping a stage are illustrative.
05
Interactive PCB & Schematic Editors
Browser prototypes for Vectis.ai
Feb – Mar 2026React · Next.js · TypeScript · SVG & canvas
2 browser editors that draw AI-generated hardware designs as a schematic and a PCB layout: selection, drag, zoom, pan, property editing, and the design file in, edited file back out.
Kept visual edits separate from engineering validity, flagging any change that still needs a backend electrical check.
Moving a component on screen is easy. Moving it without losing what it is connected to is the whole job.
1 application for the whole company site, live from November 2026: routing, server-side rendering and data-driven pages for services, specialists and case studies.
REST connections to supporting services, with enquiries wired into CRM and automation workflows.
May – Jun 2026Account scoring · Profile mapping · Lead routing
253 exhibitors and ~700 profiles ranked by strategic fit into 35 priority accounts, including Jaguar Land Rover, STMicroelectronics and Medtronic Digital Surgery.
Helped build the AI layer the team queried on the floor, inside an eight-week GTM programme.
AI content at scale, and sales research in under 30 seconds
Sep – Oct 2025n8n · Python · Web scraping · RSS · LLMs · Image generation
300+ assets a week and 780+ hours saved a month, from scraping and tone logic to visuals and delivery; the LinkedIn system reported +5,464% impressions in its first month.
880+ hours saved a month by the Personal Research Engine: 300+ reports, each under 30 seconds.
He does not treat problems as isolated technical exercises, but as interconnected systems involving trade-offs, human behaviour, and long-term consequences. … Individuals with his combination of discipline, judgment, and learning velocity tend to compound rapidly over time.
Wael Elrifai (Oxon)Senior Technology Executive, BCS FellowOn the Council →
I have seen him hold his own while working alongside engineers from top global institutions, not by asserting himself loudly, but through competence and reliability. Sankitha combines speed, humility, resilience, and technical depth in a way that is uncommon.
Aamar HussainGlobal GenAI Tech Lead, Google; formerly Director, MicrosoftOn the Council →
His curiosity is very precise. He does not ask broad or comforting questions. He challenges definitions. … He will not just keep up in a demanding technical environment. He will raise its standard.
Ruwan IddamalgodaA-Level Computer Science; Cambridge Most Influential Teacher 2017
10Volunteering & outside work
700+ hours for other people.
2023 – Present
Co-instructor, Magic Box Mathematics Programme
2,000+ students, 400+ hours: proof-based problem solving, peer marking and a lot of patience.
2023 – Present
Volunteer Web Developer
Free websites and tools for schools, teachers and small community groups.
2022 – 2023
Student Volunteer, Friends of the Earth
300+ hours of agroecology workshops, campaigning and habitat restoration.
2023 – 2025
Volunteer Chess Mentor, Ananda College
Weekly coaching for beginners and younger players: calculation, combinations and endgames.
