Mally Kisubi · London
Operations and delivery lead.I run client SLAs and build the AI tooling behind them.
Now: Regional Operations Team Lead, Chainlabs · 9 years in operations
- 500+organisations onboarded, Starknet grants
- 43people in 25+ countries, three delivery leads
- 275entities in August 2026 against a floor of 200
- $100M+treasury supported, Starknet Foundation
- What I can do
- How I work
- What I build with
Three rings round one operator. Hover an electron, click it to go there.
Selected work
Four decisions, with the numbers.
Output up: 275 entities a month against a 200 floor
After client complaints about volume: per-client SLAs, a weekly cadence, hiring across 25+ countries. 1,200+ traces that month.
Read the caseThree consoles took the vertical off Google Sheets
Ad hoc intake, cost against output, and each freelancer’s own queue, on Postgres and Node.
Read the caseGrants operations for 500+ organisations
Seed, Growth and Propulsion grants, all still open. First AML/KYB framework; cycle times cut by 33%.
Read the caseFirst UK operations hire, then Asana Champion
Set up the London office, then improved how every team ran its projects. 150+ subscriptions, €3M in annual billings, in one register.
Read the caseInside the builds
The Regional Exchange vertical, layer by layer.
Foundation first. Built with three client delivery leads and the engineering team.
Chainlabs Regional Exchange
Production · January 2026 to now
-
- Built
- The hiring pipeline, the screening assessment researchers are hired against, the weekly delivery meeting and 1:1s with the leads.
-
- Built
- The payout pipeline that pays the whole network, the crypto payment rails around it (custody addresses, invoice metadata, withdrawal-ageing digests), and the fee tracker that reconciles platform and on-chain fees across every contractor and client sheet.
-
- Built
- The Postgres data layer under contractor sheets and all four client sheets.
-
- Built
- The Node backend, Python jobs, and the migration off Google Apps Script onto the new stack.
-
- Built
- The helper chat bot the 43-person network runs on, about 17,000 lines of Node; the ad hoc, cost and freelancer consoles that replaced the shared sheets; and the client-facing portal and API feed: per-client isolation, item tracking to delivery, evidence attached.
-
- Built
- Self-monitoring so a database outage pages someone, the quality-control layer, and the evidence-retention rule so evidence survives a contributor deleting their own files.
AI I built and run
Production AI, with a budget and a gate.
At Chainlabs, on the Node, Postgres and pgvector stack the vertical already runs, at no new running cost.
A gateway with a kill switch
Three providers behind one interface. Every call priced into a ledger; a budget trip holds until a person clears it.
Hybrid-search RAG
Vector and full-text search merged by rank fusion, split by audience so ops-only text never reaches a freelancer.
Evals with a regression gate
A prompt change cannot merge if the pass rate drops against the saved baseline.
Measured on public data. nDCG@10 on BEIR FiQA, 648 queries, rank fusion over keyword search. Vectors alone scored 0.40, so measure fusion per corpus.
The repoExperience
Finance first, then the rest of the business.
- 2016GE CapitalRisk placement, Intern of the Year
- 2019Canada LifeTrainee investment accountant
- 2021The Shipowners’ ClubOperations
- 2022Parity TechnologiesCoordinator, then Project Manager
- 2024Starknet FoundationOperations Manager
- 2026ChainlabsRegional Operations Team Lead