LL.M International Business Law (Distinction) · LL.B (Hons) · Reading for the Sri Lanka Bar · Colombo
From law into systems.
I work on the legal-technology side: document workflows, correspondence automation, mandate tracking, and contract review, built so a human is always the one accountable for whatever leaves the firm. Four years of background across corporate, intellectual property, and structured-finance matters.
Case studies
What I've built.
Three system architectures from live matters, plus two longer notes on how I think about the work. Each is shown at a level of abstraction that doesn't expose any client matter; the implementation stays with the firm.
Formal documents (opinions, advice notes, demand letters, internal memoranda) drafted under firm voice and standing instructions, then routed through a human reviewer before anything is dispatched. The architecture is on the public side; the rules and templates stay private.
Open the case studyMandate facts, authority checks, document sequencing, and the next action sit in a single operating record. Source-of-truth pattern from cross-border structured-finance work.
Open the case study The document execution workflowA document only reaches signing once version, signer authority, attachments, submission route, and audit trail have all been checked. Self-hosted infrastructure.
Open the case studyDrafting examples, clause taxonomies, reviewer edits, and rejection reasons curated into reference material that AI tools can lean on without inventing facts. Out of the LL.M thesis on copyright and generative AI.
Read the note Regulation, AI accountability, and the human layerWhy the duty has to stay with the human, recent sanctions cases (Mata v. Avianca, Park v. Kim), the EU AI Act, GDPR Article 22, ABA Formal Opinion 512, and where Sri Lanka's professional-conduct apparatus fits in.
Read the noteHow the work is built
Four constraints these systems run under.
Each one is testable on the page. Stated so the claim can be checked, not taken on faith.
- Source discipline
Nothing in the output that isn't on the record.
The system never makes up a figure, an authority, a right, or a governing law. If it isn't on the matter record, the draft flags the gap in brackets. The reviewer fills the bracket or kills the draft.
An old habit from legal practice. It is what separates settled advice from speculation.
- Reviewer authority
Every output passes through a human signature.
Nothing leaves the firm without a named reviewer behind it. The system's job is to surface what the reviewer needs to see, in the order they need to see it, never to step in front of them.
Where the AI helps, it helps the reviewer think. It does not replace the thinking.
- Execution control
No dispatch until version, signer, and trail are locked.
Nothing reaches signature until the version is locked, the attachments are complete, the signer authority is verified, and the audit trail is being kept. A confident-looking draft is not the same as a document that should leave the firm.
The signing infrastructure is self-hosted, because final execution should not sit behind someone else's product.
- Public boundary
Architecture on the web. Implementation inside the firm.
The pages here show how decisions get made. Client matters, implementation detail, and the live operating register do not appear on the open site. Those are walked through inside a proper review setting.
The site is the demonstration. The firm is where the work happens.
Get in touch.
Email is the fastest route. I keep reply windows short, and any walkthrough at firm-level detail happens off the public site by design.