Ikbar Faiz

AI Engineer — Agentic Systems & Backend Infrastructure

Man, I just want to live as a farmer in the Alps with my donkey. Unfortunately, my agents won't allow that until they become the Terminator to wash me up. Wanna know about me? Just scroll down.

Ikbar Faiz
Production Experience
Software Engineer — Agentic LLM System PT Aviana Sinar Abadi Tbk PT Aviana Sinar Abadi Tbk (IRSX)
[TypeScript] [Node.js] [MCP Server] [PostgreSQL] [Playwright] [SSE] 
[ User Intent ]
       │
       ▼

[ Planner LLM ] ──── (JSON-RPC 2.0) ────► [ In-House MCP Server ]
       │                                           │
       ├─► Parallel Dispatch (≤8 Tools) ◄──────────┘
       │   (Flights, Rail, Top-Up, Catalog)
       │
       ▼

[ Server-Side Verification ] ───► [ HMAC-SHA256 & Idempotency Gate ]
       │
       ▼ (Server-Sent Events)

[ Client Result Stream ] (sub-1s resolve)
                
[ User Intent ]
       │
       ▼
[ Planner LLM ]
       │
       ├─► (JSON-RPC 2.0)
       │   [ In-House MCP Server ]
       │   └─► Dispatch (≤8 Tools)
       │
       ▼
[ Server Verification ]
       │
       ├─► [ HMAC-SHA256 Gate ]
       │
       ▼ (Server-Sent Events)
[ Client Result Stream ]
  • Intent-First Agent Backend: Engineered the agent chat backend for an intent-first AI assistant across 10 live tools (trains, flights, catalog, quota top-up, weather, stocks).
  • In-House MCP Server: Built a Model Context Protocol server (JSON-RPC 2.0) sharing tool definitions between in-house chat loop and external agents with zero duplication; verified full tool round-trip in 307ms.
  • Tool Orchestration Loop: Dispatches up to 8 parallel tool calls under a 75-second budget, shrinking per-call timeouts dynamically and streaming UI result cards over SSE.
  • Adversarial & Identity Hardening: Defended device-scoped identities against 4 cross-device attack paths using verified HMAC-SHA256 claims; isolated payment actions with idempotency keys and confirmation tokens against prompt-injection attacks.
  • Latency & Concurrency: Optimized Playwright browser concurrency for scraping tools, reducing worst-case latency from 19.2s down to 4.4s while bounding memory (+14MB for 10 concurrent sessions).
  • Rule-Based Personalization: Engineered an 18-table PostgreSQL engine deriving 6 evidence-scored trait families via statistical heuristics rather than ungrounded LLM inference.
Software Engineer — DevSecOps / Security Engineering PT Aviana Sinar Abadi Tbk PT Aviana Sinar Abadi Tbk (IRSX)
[GitLab CI/CD] [Docker Swarm] [Semgrep] [OSV-Scanner] [Policy Gating]
  • Centralized Security Pipeline: Designed and deployed a modular GitLab CI/CD security pipeline with SAST, dependency scanning, and policy gating consumed via a 3-line include.
  • Fail-Closed Policy Gating: Built a Python policy gate parsing raw scanner report payloads rather than job exit codes; unreadable or missing reports fail closed.
  • Supply-Chain Incident Patch: Patched an active npm malware bypass during the August 2026 supply-chain incident (keyv / cacheable), adding mandatory blocking for unrated MAL- advisories verified against OSV records.
  • Alert Noise Reduction: Collapsed 46 raw scanner alerts into 3 actionable root causes while maintaining human-in-the-loop merge controls and advisory LLM reviews.
Publications & Research
Stuck at the Same Wall: Why Biological Brains and Artificial Intelligence Hit the Same Explanatory Limit PhilArchive [FAISAT-2] ↗
Philosophy of Mind • Artificial Consciousness • Epistemology
Analyzes the "Loop of Consciousness"—the persistent epistemological boundary where mechanistic explanations of cognition recur back to the subjective observer. Maps self-referential limits using parallels to Gödelian incompleteness, computational irreducibility, and demonstrates why structural transparency in artificial neural networks does not yield phenomenal accessibility.
Kinematical and Photometric Deconstruction of the UGC 694-IC 412 System: Evidence for a Line-of-Sight Projection arXiv:2512.22661 [astro-ph.GA] ↗
Extragalactic Astrophysics • SDSS DR16 Kinematics • Differential Photometry
Re-evaluated the assumed gravitational interaction of galaxy pair UGC 694 and IC 412 using SDSS DR16 radial velocity data. Demonstrated an 8,372 km/s velocity offset corresponding to a massive 123.5 Mpc spatial separation, proving the perceived tidal bridge is an optical line-of-sight projection rather than a physical encounter.
Data Systems & Observational Pipelines
NASA NASA IRIS — Archive Ingestion & Validation Framework
github.com/arsenelupin14/iris-solar-uvdata ↗
[Python] [FITS] [Pytest (83 Tests)] [LMSAL Data] 
[ Raw Observational Telemetry ] (FITS / Parquet)
       │
       ▼

[ Deterministic Ingestion Engine ]
       │
       ├─► Near-Real-Time Data Segregation
       ├─► Header Metadata & Coordinate Parsing
       └─► FITS Checksum & Integrity Gate
       │
       ▼

[ Science-Ready Artifacts ] (83 Passing Pytest Suites)
                
[ Raw Telemetry ] (FITS/Parquet)
       │
       ▼
[ Ingestion Engine ]
       ├─► NRT Data Segregation
       ├─► Header & Coordinate Parse
       └─► FITS Checksum Gate
       │
       ▼
[ Science-Ready Artifacts ]
  (83 Passing Pytest Suites)

Engineered an ingestion and verification pipeline for Lockheed Martin Solar & Astrophysics Laboratory (LMSAL) IRIS Level 2 observation products. Segregates Near-Real-Time vs. regular science telemetry, validates FITS checksums, and produces audit-ready screening reports with 83 passing test suites.

NASA T CrB Observational Batch Pipeline
github.com/arsenelupin14/t-crb-project ↗
[Pandas] [NumPy] [Parquet] [Spearman: 0.883]

Batch pipeline ingesting and validating 174,872 astronomical records from multiple independent public observatories and NASA high-energy databases. Implemented rolling median/MAD smoothing and verified cross-source consistency with Spearman correlation of 0.883 across overlapping observation windows.