Autonomous worker fleets, quantitative decision engines, and resilient enterprise platforms engineered to run reliably under pressure.
We are KHANSTER.NET: an engineering practice designing and shipping production AI systems, autonomous agent fleets, and AI-powered operational software.
Inspired by the functional resilience of Starfleet systems, we build software designed to run reliably under pressure. No research theater, no speculative abstractions, and no synthetic marketing metrics. From sub-50ms trading decision engines to air-gapped financial reconciliation platforms, we build systems that solve tangible operational bottlenecks.
Every system we deploy is grounded in deterministic safeguards, clean system seams, and verifiable audit trails that perform in production from day one.
Pragmatic AI roadmaps tied directly to your operational bottlenecks. We evaluate architecture feasibility, model selection, and unit economics before writing code.
Production integration of LLMs, agentic loops, and multimodal systems into existing workflows. We build grounded RAG pipelines, deterministic guards, and structured data outputs.
Operational modernization centered on autonomous agent workflows and automated execution. We help internal engineering teams adopt AI tooling, establish governance, and scale systems safely.
Deploy AI systems with verifiable safety, strict data isolation, and auditable outputs. We implement boundary testing, hallucination guards, and compliance baselines for regulatory standards like the EU AI Act.
Unified operations and sales execution workstation connecting enterprise pipeline tracking with RAID delivery governance. Features automated weekly ransomware threat intelligence ingestion, server-side entity deduplication, cryptographic HMAC client magic links for one-click RAID sign-off, and automated executive portfolio digests. Operates on a dual-stack PostgREST engine with vector embeddings, zero client data leakage, and rigorous UAT gate cutovers.
LIVE
FLAGSHIP
Enterprise partner lifecycle portal for Microdium’s national cybersecurity reseller network. Implements a 5-step onboarding wizard with in-browser digital signature capture, automated PDF agreement generation, authorized partner certificates, deal registration, and quotation workflows. Features one-click magic email approvals for administrators, strict multi-role authorization (Partner/Staff/Admin), Drive-synced marketing collateral distribution, and complete deal confidentiality.
LIVE
Bounded agentic AI assistant developed for Microdium Partner Hub to handle administrative and operational workloads, engineered as the fleet’s reusable architecture for administrative copilots across enterprise platforms (including Orbit). Hooked up to the DeepSeek-V4 model, Alia operates under established four-tier structural guardrails: scoped read tools query live Postgres data with zero hallucination, reversible direct actions, server-persisted confirmation cards requiring explicit human approval before any mutation executes, and fail-closed preparation for sensitive outbound actions.
LIVE
FLAGSHIP
Enterprise standing month-end close toolbox and multi-engine reconciliation platform. Executes cross-system matching across Oracle Fusion AR/GL and SAP Ariba receipts vs Fusion GL accruals (GR/IR) with cell-for-cell audit parity. Features a 9-category deterministic root-cause classifier, aging buckets, automated 4-sheet SOX close packs, and 100% in-browser airgap privacy with zero cloud exposure, backed by a cryptographic client distribution portal.
LIVE
Autonomous crude palm oil futures decision engine and paper trading terminal powered by TypeSafe Jev System One. Ingests multi-timeframe TradingView webhooks across 1m to 30m sleeves, executing sub-50ms non-autoregressive setup grading and risk modeling. Features real-time SSE streaming, interactive Lightweight Charts, multi-session auto-flattening risk gates, and confidence-collapse chop protection in volatile market conditions.
LIVE
Continuous vulnerability assessment platform for managed security providers, now live and running real client scans. Seven deterministic engines execute a five-phase pipeline under locked safety profiles, feeding a triage layer that de-duplicates across engines, scores each finding with its own CVSS 3.1 vector, and maps impact to OWASP, ISO 27001, PDPA and Bank Negara RMiT.
PROTOTYPE
Sovereign AI travel companion and tourism intelligence platform prototype engineered for the Malaysian Technology Development Corporation (MTDC) national pitch. Combines an 8-view single-page application with conversational multi-day itinerary generation, localized destination discovery, operator directory listings, and bilingual EN/BM voice synthesis, backed by a 4-tier revenue model and a self-hosted, foreign-API-independent AI architecture.
LIVE
B2B cybersecurity prospecting workstation and intelligence CRM covering all 1,078 Bursa Malaysia equities (Main & ACE markets, RM 2.13T cap). Merges daily fundamentals, multi-layer Exa leadership discovery, and Cyber Security Act 2024 (Act 854) NCII compliance tier mapping to isolate verified IT decision-makers (CIO/CISO), featuring one-click outreach hook generation and direct pipeline handoff to Orbit CRM.
LIVE
FLAGSHIP
Futures trading academy platform and marketing funnel managing 20-50 student cohorts across multi-coach academies. Bridges public registration to coach approval, server-tracked video masterclass engagement (≥80% completion gate), and structured student readiness tracks. Includes an instant A-B-C-D setup calculator, SOP reference guide, journal adherence scoring, Telegram fan-out broadcasts, WhatsApp deep-link hand-offs, and priority Q&A threading in full EN/BM bilingual UI.
LIVE
Multi-tenant corporate visitor management platform engineered for facility gate security and strict tenant data isolation. Features a multi-step self-service mobile kiosk with live camera photo capture, anti-duplicate check-in guards, returning-visitor privacy-masked lookups, and encrypted QR zone passes. Dispatches instant arrival notifications with inline visitor photos to corporate hosts, backed by per-company analytics dashboards.
BENCHMARK
Semantic search without a vector database. Instead of embedding anything, we put TypeSafe Jev in front of the index we already had: the model resolves the question into a typed decision, the existing local index answers, and a generative LLM is called only when an answer actually needs drafting. Measured across 324 questions on the Partner Hub’s real production index (3,819 chunks, 133 documents). Both directions are statistically significant. Where the asker uses the document’s own words, keyword search wins 65% to 33% (p<0.0001), where a model makes retrieval worse. Where the wording does not match, the model wins 12.1% to 1.5% (p=0.046), and on questions sharing no vocabulary with the right document keyword search scores 0%. The two cross over at roughly 0.45 overlap: gate there, answer the rest locally and free. No embeddings, no vector store, none needed.
LIVE
B2B sales intelligence CRM compiling 11,240+ verified halal-certified Malaysian enterprises scraped directly from official JAKIM directories. Enriches company profiles with ownership classification, contactability metrics, certificate expiration forecasting, and multi-state category distribution. Built with strict OWASP CSP headers, custom JSON data-islands, delegated event routing, and an interactive team pipeline board featuring automated 30-day activity pulse tracking.
LIVE
Searchable technical talent directory presenting 25 software engineering graduates from 42 Malaysia (Sunway campus), built spec-to-production in 3h 44m via an autonomous agent fleet. Features deterministic skill-overlap and term-frequency job description matching, canvas-rendered capability radars, side-by-side candidate comparison, and zero-PII client serialization with strict administrative publish gating and privacy protections.
LIVE
Retail front-of-house and ordering web app for Homebaker@Clover’s micro-kiosk and central kitchen network. Combines a custom static frontend featuring Day/Dusk mood switching, interactive Box of 4/6 builder, and kiosk radar with a native headless Ghost CMS backend on Besen VPS. Built for the founder to publish journal essays, update roll prices, and broadcast operating notices straight from phone or laptop without code changes.
BENCHMARK
A 3D rail-shooter benchmark and spatial reasoning shootout comparing frontier LLM code generation. Evaluates GLM 5.3-max against Gemini 4 Argon on an identical 64-line specification prompt requiring procedural Three.js flight mechanics, WebAudio synthesis, and skyline landmarks with zero external assets. Features a 4-engine interactive browser sandbox, automated headless Chrome test harness analysis, and Malaysian-themed engine evolution.
BENCHMARK
A real-world engineering benchmark comparing lone coding assistants against bounded multi-agent fleets operating in real time. Shows how isolating capabilities across specialized foundation models in the Herdr terminal multiplexer (Kilo on backend code, Hermes on browser DevTools, Antigravity on synthesis) autonomously resolved a silent cross-origin OAuth deadlock without human log relaying or context bloat.
LIVE
Unified infrastructure observability and defensive telemetry cockpit providing centralized visibility across distributed production environments. Combines high-fidelity health monitoring with plain-language security event translation and tenant-isolated reporting portals. Tracks real-time process uptime, resource consumption trends, and anomalous traffic patterns through cryptographically gated administrative sessions and automated operational alerts to maintain continuous availability.
LIVE
Editorial workbench and creator studio built for engineering articles and technical analysis. Aggregates post candidates from fleet logs, commit histories, and operational incident records directly into Problem-Agitate-Solution drafts. Features a LinkedIn feed simulator with fold emulation, Unicode formatting toolbar, a word-level Voice Gap diff engine tracking draft drift against published copy, and a candidate-scoped Hermes copilot.
LIVE
LCARS mission control cockpit and fleet operations console governing a four-agent autonomous Hermes crew. Houses specialized roles across orchestration, live intelligence, systems engineering, and client-facing editorial drafting. Features real-time bridge telemetry, streaming subprocess chat, automated mission dispatch, inter-agent handoffs, cron scheduling, and server-sent event terminal access through a zero-dependency Python service behind cookie-gated edge authentication.
Corporate web presence for Melewar Integrated Technology: responsive static architecture built with Astro and Tailwind CSS, delivering fast edge caching and clean accessibility.
Decision-support simulation tool for an Oil & Gas consortium: models vessel turnaround cycles, supply-chain bottlenecks, boil-off gas rates, and contract yields under fluctuating market variables.
Distributed multi-agent autonomous engineering fleet: CEO/worker agent hierarchy operating across VPS nodes with a 3-layer shared memory stack coordinated via Model Context Protocol.
Autonomous Android device-fleet orchestration platform: zero-touch bulk device provisioning, distributed command execution, telemetry capture, and remote policy enforcement across hardware fleets.
Terminal-style Bursa Malaysia equities screener and portfolio trade journal with real-time technical filters and offline trade analytics.
Executive governance and operations workstation streamlining qualification reviews, committee approval workflows, and audit-grade disbursement tracking.
Always-on WhatsApp agent gateway and private concierge service delivering automated reply drafting with strict human verification gates.
Local-first workstation generating structured enterprise sales proposals with deterministic commercial validation and multi-format document export.
Zero-write corporate mail intelligence overlay delivering scheduled executive Telegram digests and protected conversational query interfaces.
Ready to begin your AI transformation? Hailing frequencies are open. Initiate contact and our team will respond within one Federation standard day.