01 No baseline no signal
A system you can't measure is a system you can't fix. Most projects launch with no agreed definition of what working means, so nobody can tell improvement from noise.
05 / Handover
We hand over something you can run.
You own the code and the system, documented and handed over so whoever runs it day to day actually can. We train them, and we stay reachable.
Custom Software AI Agents Automation AI Reliability Audits
For founders, SMEs & growing teams
- 14named projects
- 4built by SerSan
- 1PhD, applied maths
The build-to-payoff gap
Most projects don't fail on the day they go live. They fail two months after.
The demo worked. Everyone nodded. Then real volume arrived, the edge cases piled up, someone quietly went back to doing it by hand, and no-one could say which of the seven changes made last month broke it.
02 No trail no debugging
When the system does the wrong thing, you need to know which step failed. Without a recorded trail, every incident turns into archaeology.
03 No boundaries no trust
Access to data, money and customers without a permission model becomes a liability the first time something acts without asking.
What production-grade means, when it's needed
Three things every SerSan system ships with,
scaled to what it actually carries.
Not a list of compliance buzzwords. Each one is scaled to the system. These are artifacts you can ask to see in any scoping call.
01 eval baseline
Proof it still works after every change.
Versioned test cases, sized to the system: a handful for a workflow tool, a full regression set for a platform. You prove it, instead of hoping.
02 trace propagation
Traceable from input to action.
When something goes wrong, the answer is in the record: what came in, what it decided, what it did, who reviewed it. Not Slack archaeology.
03 guardrail clamp
Boundaries before features.
What the system can reach — data, accounts, actions — is agreed before the first feature. The default answer to an unscoped action is no.
We do not claim compliance certifications we don't hold.
We do build systems that pass them.
Selected work
Engineering you can name.
14 projects across FinTech, Agritech, Real Estate, Industrial, Healthcare, Public Sector, Aerospace and Energy. SerSan builds and prior senior delivery, each one labelled for what it is.
SerSan delivery · FinTech · Trading Education · RAG Mentor · Community · CRM
SerSan delivery · FinTech · Quant · LLM Agents · Real-Time Data
SerSan delivery · Real Estate · Brand Site · Motion Design · Lead Generation
SerSan delivery · Agritech · Renewables · IoT · Predictive Maintenance · SCADA
Prior experience — Michele Sanna at Revolut · FinTech · Fraud Detection · Real-Time Inference
Prior experience — Michele Sanna at J.P. Morgan · FinTech · Treasury · Credit Risk · Aerospace Research
Prior experience — Michele Sanna at Deloitte · Retail Operations · Time-Series ML · Supply Chain
Prior experience — Michele Sanna, research grant · MedTech · Computer Vision · Oncology Research
What SerSan builds
Four services. One discipline.
Start with the problem. Build the smallest useful solution. Scale what works.
Service 01
Custom Software & Platforms
The software your business runs on, built to fit.
Typical build includes
- Internal tools and management systems
- Customer portals and web applications
- SaaS products and academy/LMS platforms
- Integrations across the tools you use
Solves
- The tools you run on don't talk to each other.
- You need software built properly, without hiring a team.
Service 02
Workflow Automation
Start with the process that eats the most time.
Typical build includes
- Repetitive back-office work, handed off
- Documents read, checked and filed
- Leads, orders and CRM data kept in sync
- Nothing lost: failures surfaced and retried
Solves
- Your team retypes the same data every week.
Service 03
AI Features & Reliability
AI that holds up once real customers use it.
Typical build includes
- Assistants and copilots on your own content
- Agents that act, with the limits you set
- Tested against real cases before release
- Monitored live: cost, quality, accuracy
Solves
- The AI demo impressed everyone, then stalled.
Service 04
Technical Audits & Architecture
Find what should not be built, before code becomes debt.
Typical build includes
- Systems and process audit: where it hurts
- Build vs. buy vs. don't-build call
- Where AI would actually pay off
- Target architecture, sequenced and costed
Solves
- You know what's broken, not what to build first.
- You need readiness before a board, customer, or regulator.
Founder-led software, AI & automation
Built by the people who build and run it.
You work with the builders. Every engagement has a named commercial owner and a named technical owner. Accountability stays senior, whoever else contributes.
- 01 / 04
Owns scoping, proposals, pricing, client communication, and engagement structure from first call to handover. He turns ambiguous business problems into clear technical briefs, fixed scopes, and accountable delivery.
- USAAI Certified AI Consultant (CAIC)
- Dual Master's: MBA + AI in Business
- 02 / 04
The named technical owner on an engagement — from a single automation to a platform in production. Owns the build, the data path, and the AI layer where it earns its place.
- PhD Applied Mathematics, LSE
- 8 years senior delivery at tier-1 institutions, pre-SerSan
Previously- J.P. Morgan
- Revolut
- Deloitte
- Brevan Howard
- Accenture
- 03 / 04
Owns the front-end, WebGPU and motion layer of what SerSan ships, this website included, and the AI agents built on the Microsoft stack. Also owns the studio's creative direction: the SerSan logo, its motion-graphics ads and video editing.
- BSc Computer Science, Università di Camerino (2026)
- Thesis: architecture and security of OS-level agentic systems (MCP, multi-agent, Zero-Trust)
- Built this site: WebGPU rendering, 3D and motion
- 04 / 04
Designs and builds internal systems end to end — requirements, data model, interface, delivery. Shipped a maintenance and inventory management system for a resort operator, from formal requirements analysis through to the running software.
- BSc Computer Science, Università di Camerino
- Published: Knowledge Graphs as a Semantic Layer for Understanding Robotic Video
Honest about fit
We are honest about who we work with.
Good fit: Something in the business costs you time or money today.
Not a fit: You want a chatbot gimmick for a press release.
Good fit: You can explain the workflow, even if not the technology.
Not a fit: No one can explain what problem is actually being solved.
Good fit: You'd rather fix one real problem than buy another licence.
Not a fit: You want a fixed price before anyone has defined the work.
Good fit: You want it built properly, whether or not you're regulated.
Not a fit: You want to skip compliance to ship faster.
Good fit: Someone on your side can decide and unblock us, technical or not.
Not a fit: Nobody can actually decide, so nothing gets signed off.
Good fit: You can give us the people, data or access to understand it.
Not a fit: You won't give us the access or feedback the work needs.
Start with the problem
Bring us the problem. We'll bring the plan.
One workflow, one product idea, one system that keeps breaking, or something your team still does manually. Tell us what's happening. We'll tell you what we'd do next.
what_you_get.tsfixed scope
// after your briefconst whatYouGet = { readBy: "a founder", price: "free", deliverable: "a written next step", verdict: "build | harden | stop",};
Reply within one business day