Scale & Leadership
Standard operating procedures, shift scheduling, and KPI benchmarks for 50+ operators across multi-tier QA.
Available for AI Operations Leadership & Agent Architecture
Ahsan Firdaus · AI Operations Lead & Agentic Systems Architect
I lead a 50+ person team delivering ground-truth perception data for Xiaomi autonomous vehicles.
Delivered 7,870 multimodal datasets at 100% post-revision accuracy in March 2026. I bridge client engineering specifications with high-volume data operations, designing multi-tier QA protocols and resilient AI agent loops.
- 50+ Annotators & QA Specialists Led
- 7,870 Datasets Delivered in March 2026
- 100% Client Accuracy Post-Revision
About
Trained in informatics, proven in scaled operations.
- Location
- Depok, West Java, Indonesia (WIB · UTC+7)
- Current role
- Team Leader, Autonomous Vehicle Project
- Organization
- KUPU ID — PT Dalligent Solusi Indonesia (Sinarmas Group)
- Key client
- Xiaomi autonomous vehicle program
- Education
- M.S. & B.S. in Informatics, Universitas Gunadarma (GPA 3.82 / 3.80)
- Availability
- Open to Full-Time Leadership & Strategic Engagements
I turn client engineering specifications into high-throughput production reality for a 50+ person data annotation and QA team.
I hold a Bachelor of Informatics (GPA 3.80) and a Master of Software Information System (GPA 3.82) from Universitas Gunadarma. My academic research focused on machine learning architectures, financial time-series modeling, and resilient data pipelines.
At KUPU ID — PT Dalligent Solusi Indonesia (Sinarmas Group), I lead annotators and QA specialists who produce multimodal perception data for Xiaomi's autonomous vehicle program. I direct daily shift allocations, write multi-tier QA protocols, and bridge communication between client engineering teams and front-line operators. In March 2026, our team delivered 7,870 datasets at 100% post-revision accuracy.
Alongside production data operations, I design and prototype autonomous agent loops. I combine large language models, structured schemas, operational databases, and mandatory human review checkpoints into resilient automation workflows that hold up in enterprise environments.
Agentic Loops
Multi-agent delegation, schema-enforced tool calling, and human review gates for mission-critical workflows.
Multimodal Sensor QA
Ground-truth verification for 2D camera images and 4D LiDAR temporal point clouds against sensor tolerances.
Systems Engineering
Custom JavaScript, backend plugins, and FetchXML automations for Microsoft Dynamics 365.
Work
Selected projects.
Autonomous agent architectures, autonomous vehicle data operations, and production software systems.
Ahsan Labs · Active Architecture
A testing ground for autonomous agent loops and tool orchestration
Ahsan Labs evaluates how multi-agent delegation, structured schema enforcement, and human checkpoints automate enterprise operations while preventing hallucinations and unhandled exceptions.
KUPU ID — PT Dalligent Solusi Indonesia (Sinarmas Group) · Xiaomi AV
Perception data operations for Xiaomi autonomous vehicles
Directing a 50+ person team of annotators and QA specialists working across 2D camera streams, 3D cuboid projections, and 4D LiDAR point clouds. Authored standard operating procedures, resolved sensor edge cases, and sustained zero-defect delivery standards.
7,870Datasets delivered in March 2026
100%Client accuracy post-revision
50+Annotators & QA specialists led
More projects
AI creative tooling
PPV
A guided prompt builder for AI image and video generation. It turns loose creative intent into structured, model-ready prompts through a step-by-step flow, so consistent results no longer depend on memorizing model-specific syntax.
Image + videoOne guided flow for both generation modalities
Privacy-first web utility
San Doc Convert
A client-side document and image conversion toolkit built with Next.js. Converts formats (PDF, JPG, PNG, WebP), merges, extracts, and rotates files directly in the browser—guaranteeing 100% on-device processing with zero data uploaded to external servers.
100% on-deviceZero server transmission & immediate session purge
Client-side security tool
Watermark v1
In-browser batch watermarking engine for confidential identity verification (passports, national IDs, certificates). Supports multi-image batching, opacity adjustment, rotation, and ZIP export without any cloud dependencies.
10-image batchLocal canvas rendering with bilingual ID/EN interface
Machine learning
Credit card fraud detection
Fraud cases are rare in transactional training datasets, causing standard classifiers to miss anomalies. Rebalanced distributions with SMOTE resampling and tuned Random Forest and XGBoost models.
0.15 to 0.91Recall increase on fraud instances
Computer vision
Vehicle counting from traffic video
Trained and fine-tuned a YOLO object detection model via PyTorch transfer learning on 1,667 road traffic images to count multi-class vehicles across video frames.
90%Detection precision across test footage
Enterprise automation
Dynamics 365 workflow automation
At Dinamika Teknologi Informasi, customized Dynamics 365 LMS and Middle Office modules using JavaScript, backend plugins, and FetchXML to automate approval workflows and data reconciliation through UAT sign-off.
Requirements to UATDelivered end to end for enterprise clients
Product management & health-tech
Day Healthy platform ecosystem
Engineered product specifications, user journeys, and technical roadmaps at Binar Academy for an integrated health-tech solution spanning automated consultations, claims processing, and pharmacy logistics.
4 core modulesAI chatbot, insurance verification, pharmacy delivery, and admin portal
Bangkit capstone
Daily Cloud, mental wellness journaling app
Android mobile application that detects facial emotion using TensorFlow mobile inference and evaluates the sentiment of journal entries using NLP to track wellness trends over time.
95%Facial emotion classification accuracy
Exploring
Core principles in agentic operations.
Five operational themes from agent testing, paired with an interactive terminal simulator modeling production tasks.
01Autonomous AI agents
How an agent breaks a goal into steps, tracks execution state, calls designated tools, and delegates tasks to subagents within strict security boundaries.
02Agentic loops vs. deterministic scripts
Static if/else code breaks when multimodal input drifts. An LLM agent loop self-corrects, retries failed tool calls, and recovers without crashing production.
03AI-assisted QA rubrics
Standard operating procedures where models perform initial coarse-pass triage on high-volume datasets, routing edge cases to experienced human reviewers.
04Workflow automation and RPA
Integrating models directly into operational databases, REST APIs, and team dashboards to remove repetitive data entry and status tracking.
05Human-in-the-loop governance
Operational teams where technical leaders establish confidence thresholds, audit model decisions, refine prompts, and monitor throughput velocity.
00:00INITSelect a scenario above and press "Run scenario".
Simulation runs against test schemas with synthetic identifiers.
Experience
Leadership and engineering track record.
Promoted rapidly based on operational precision, system ownership, and cross-functional leadership.
KUPU ID — PT Dalligent Solusi Indonesia (Sinarmas Group)
Annotator to Team Lead in three months.
- Data Annotator
- Quality Assurance
- Team Lead
Team Leader, Autonomous Vehicle Project
KUPU ID — PT Dalligent Solusi Indonesia (Sinarmas Group)
Directing a 50+ person team of data annotators and QA specialists delivering perception datasets for Xiaomi autonomous vehicle models. Reporting directly to the Project Manager with 10 direct reports.
- Delivered 7,870 datasets in March 2026 with 100% accuracy after revision.
- Coordinated the workflow from 2D camera labeling through 3D cuboid projection to 4D LiDAR temporal QA.
- Run daily shift scheduling, KPI audits, and edge-case rulings to stay above client benchmarks.
- Advanced from Annotator to QA to Team Lead within three months.
Course Assistant and IT Helpdesk
Lembaga Komputerisasi, Universitas Gunadarma
Taught 9-week lab courses on database management, SQL, and ERP systems (Adempiere, Microsoft Dynamics NAV) to 200+ students across 5+ classes, and maintained virtualization infrastructure.
- Maintained 5+ VMware vSphere hosts and 100+ workstations on Microsoft Active Directory, cutting downtime by 10%.
- Mentored students through relational database normalization, complex SQL queries, and ERP transaction processing.
Junior Technical Consultant
Dinamika Teknologi Informasi
Customized Microsoft Dynamics 365 LMS and Middle Office modules for enterprise clients, from requirements analysis through technical customization to User Acceptance Testing.
- Engineered custom modules with JavaScript, backend plugins, and FetchXML to automate approval workflows and data reconciliation.
- Managed technical deployments and data verification during enterprise UAT migrations.
Machine Learning Intern
Encryptix
Developed and benchmarked machine learning models for fraud detection and text classification. Increased recall on imbalanced fraud transactions from 0.15 to 0.91 using SMOTE resampling with Random Forest and XGBoost ensembles.
Product Management Student
Binar Academy
Collaborated within a cross-functional squad of 7 (PM, UI/UX, QA, and Engineering) to define product requirements and technical architecture for the Day Healthy platform under Agile/Scrum.
- Authored comprehensive Product Requirement Documents (PRDs), user journeys, and prioritized product roadmaps across 2-week sprints.
- Engineered feature specifications for 4 integrated modules: AI health consultation chatbot, insurance claims verification, online pharmacy ordering, and administrative dashboard.
Machine Learning Cohort, Distinction Graduate
Bangkit Academy by Google, GoTo, and Traveloka
Graduated with Distinction in machine learning. Trained in TensorFlow, computer vision pipelines, and cloud deployment. Led the ML track for Daily Cloud (95% facial emotion classification accuracy).
Education
Degrees and credentials.
Master of Software Information System
Universitas Gunadarma
Bachelor of Informatics
Universitas Gunadarma, Faculty of Industrial Technology
- Distinction GraduateBangkit Academy (Google, GoTo, Traveloka), 2023
- CCNA: Introduction to NetworksCisco, 2025
- CCNA: Switching, Routing and WirelessCisco, 2025
- TOEFL ITP, Score 597English Proficiency Certification
- 2nd Runner-up, LIGA CODING for BeginnersCompetitive Programming, 2022
Playbooks
Operational frameworks from production.
Methodologies and architectural insights drawn directly from autonomous vehicle perception pipelines and autonomous agent deployments.
Operations & Quality Engineering
Holding 100% Accuracy Across 7,870 Multimodal Datasets
How a 3-tier validation protocol keeps 50+ operators calibrated against strict client benchmarks without throughput decay.
System architecture: Annotator self-check → Automated geometric heuristic filter → Senior QA edge-case adjudication.
Proven result: Cut rework cycles by 34% while delivering 100% client acceptance compliance on 7,870 frames in March 2026.
Playbook 01 · Production SOP
Agentic Architectures
Why Deterministic Scripts Crash and How Agent Loops Recover
Handling schema drift, timestamp anomalies, and sensor dropouts using LLM-routed subagents and fallback retries.
System architecture: Schema validator → Planner with goal decomposition → Isolated tool sandbox → Human verification gate.
Proven result: 0 unhandled fatal crashes across batch sensor triage runs with automated anomaly notifications.
Playbook 02 · System Design
Perception & Sensor Systems
Spatial Bounding Alignment in 4D LiDAR Point Clouds
Maintaining millimeter-level calibration tolerances and temporal coherence across vehicle camera and roof LiDAR sensors.
System architecture: Extrinsic transformation matrices validation → Temporal bounding interpolation → Occlusion reflection checks.
Proven result: Maintained sub-2cm spatial boundary precision across high-density urban traffic sequences.
Playbook 03 · Sensor QA
Connect
Ready to scale your AI operations or automate critical workflows?
Available for technical leadership roles, AI operations consulting, and high-throughput data pipeline architecture. Based in Indonesia and collaborating with teams worldwide.
Available for Full-Time & Strategic Engagements
- Direct Email (Click to copy) ahsanfirdaus1@gmail.com
- LinkedIn Profile linkedin.com/in/ahsanfirdaus1
- GitHub Repositories github.com/ahsanfirdaus1
- Curriculum Vitae Download PDF (127 KB)
Direct response within 24 hours · Jakarta WIB (UTC+7) · Remote / Hybrid / On-site