Nasik Abbas

h9k is a new product at Moonlabor, the company PadelOS was spun out of. It is in early development.

• The idea in one line: ask your colleague's AI agent instead of interrupting them.

PadelGPT is the AI assistant inside PadelOS. It answers players in their own language, at any hour, from each club's live bookings, prices and rules.

• 15 LangGraph agents — two supervisors routing to 13 specialists — behind a player surface and an admin surface, calling 169 tools (102 admin and 67 player) on two FastMCP servers, reached in-process and over HTTP for white-label consumers.
• Tenancy enforced server-side: identity parameters are stripped from every tool schema the model sees and injected at call time from server-set configuration, so a prompt cannot widen its own scope.
• Text-to-SQL as a bounded, deterministic LangGraph StateGraph — validate, select tables, select columns, write, execute, review, at most three loops — on o3 and o4-mini against a read-only replica. It powers audience segmentation, with a controlled-vocabulary ontology and a classifier that refuses discriminatory targeting.
• Retrieval over the club knowledge base on a FAISS inner-product index of normalised embeddings, rebuilt by temp-file-then-rename so a live server never reads a half-written index.
• Measured the two admin surfaces against each other over a quarter: segment use grew 7× to 40% of admin messages at 6.3% of admin model spend. The efficient surface was the one clubs adopted, so the roadmap moved onto it.
• 33 versioned releases in six weeks, promoted main → staging → release → prod. Authored 601 of 818 commits in the repository.

Pollen tracked food from farm to shop in Saudi Arabia, so supply problems and spoilage could be caught early. It was built at Moonlabor, the company PadelOS was later spun out of.

• Device-agnostic IoT ingest: Teltonika, BiTrek and Digital Matter trackers on roaming IoT SIMs, with data arriving over MQTT through a webhook or by polling, queued, and consumed in chunks.
• Simulacrum, an IoT fleet simulator: continuous-monitoring and inter- and intra-city travel modes, a configurable fleet and routes, and route choice mined from real logistics data, so models could be tested against realistic streams around the clock.
• Anomaly triage with a human in the loop: the system flags and ranks supply-chain anomalies and recommends an action; the decision stays with a person.
• Platform: NestJS microservices on AWS EKS, React, PostgreSQL, Redis, RabbitMQ, the ELK stack and CloudWatch.

A medical-technology startup working to restore sight through a brain‑computer interface.

• Isaac-1, an MVP that lets quadriplegic users control digital devices with eye movement — the first step toward the visor the full system needs.
• Built the research team around a prospective co‑founder, a postdoctoral researcher at MIT, and grew it from there. The concept: a wearable visor and camera, a high-bandwidth wireless link, and an electrode array in the visual cortex that produces a spatial pattern of phosphenes.

An AI maths tutor for school students: photograph a question and get the solution, explained step by step.

• An LLM-based tutor for maths and science, built before ChatGPT launched.
• 1,500 paying users in six weeks, acquired without paid marketing.
• Owned the product lifecycle — discovery, product requirements, UX direction, roadmap and OKRs, go-to-market — and the finance function; raised about $2 million in cloud and software credits (AWS, Segment, Mixpanel, Intercom) to fund the build.

An app for investing in property from a few dollars.

• A marketplace for units of closed-ended real-estate funds: pooled bookings held in third-party escrow, released on full subscription, and resaleable in the app.
• 3,122 waiting-list sign-ups in under three months, with no paid acquisition.
• Stopped for want of a licence from the Securities and Exchange Commission of Pakistan (SECP).

• The reports and financial models for four TUSDEC feasibility studies under the National Strategic Programme for Acquisition of Industrial Technology — testing facilities for construction materials and technical textiles, waterless dyeing for sportswear, and digital design for light engineering: three-statement projections with scenario and risk analysis, feeding cost–benefit analysis and Planning Commission proposals.
• Field research and analysis for the cutlery-sector benchmarking study: a systematic random sample of 58 firms from 280 registered firms, a questionnaire field-tested and cut from 55 items to 41, analysis in SPSS, and findings validated in five focus groups.
• Designed the Own-Operate-Develop-Educate partnership model for PIDC's proposed cutlery common production facility: an adaptation of the World Bank's Own-Develop-Operate PPP in which a partner university owns and operates the facility, and its revenue is ring-fenced for expansion, R&D and an endowment.
• Lean transformation at a manufacturer: value-stream mapping, 5S and Kanban, and production-planning models on five years of production and sales data (4% lower cost of goods sold in year one); warehouse flow and shelving designed with closeness matrices and block plans; an online store launched at 6.2% conversion and 30% returning customers.
• Odoo ERP end to end — CRM, documents, HR, finance, purchasing, manufacturing, inventory, warehouse, point of sale and portals — with the platform chosen by Analytic Hierarchy Process alongside the executive team; and a credit scorecard system for a microfinance lender, for loans up to PKR 50,000.

• Solar-powered corrosion protection for gas-field pipelines in Sindh: installation lead on remote sites, and operator training for Pakistan Petroleum Limited.
• Brought PVsyst into design and modelling, automated equipment selection and material estimation with Excel macros, and led two engineers building design tools.
• Rebuilt the trainee programme: four and a half months to under two months, a 62.5% cut.
• The full procurement cycle for imported equipment: tender analysis, supplier quotations and negotiation, letters of credit, port clearance and delivery to site.

• Lean and TPM at a food plant moving from TQM to World Class Manufacturing: contamination-source elimination, a warehouse capacity and dwell-time study, an SOP compliance audit of the warehouse management system, and KPI monitoring with root-cause analysis.