Machine Learning Researcher
MBZUAI NLP Collaboration
Mohamed bin Zayed University of Artificial Intelligence is a graduate research university in Abu Dhabi focused on artificial intelligence.
Nov 2025 — PresentBuilt and evaluated SCOPE, which led on 6 of 8 metrics across 200 outlets, and developed annotation and modeling workflows for a harmful memes dataset of nearly 14K samples.
• Co-designed and implemented the multi-agent architecture and nine specialized analyzers for evidence-grounded media profiling, spanning article acquisition, content analysis, external evidence search, scoring, and schema-constrained outputs.
• Built the complete interactive demonstration and audit interface, surfacing article evidence, failed-fact-check signals, propaganda spans, bias components, and per-agent structured outputs.
Machine Learning Researcher
MCML / TUM Collaboration
The Munich Center for Machine Learning is a German research center bringing together machine learning groups, including teams at TUM.
Mar 2026 — Jul 2026Contributed Kazakh data and evaluation to an 18-language math reasoning benchmark, validating 500 problems and auditing model outputs and reasoning traces.
• Kazakh-language contributor for an 18-language mathematical-reasoning benchmark; validated 500 problems across four difficulty tiers for semantic fidelity, mathematical terminology, and LaTeX integrity.
• Audited Kazakh LLM outputs and reasoning traces across the benchmark evaluation, separating final-answer correctness from instruction following, reasoning coherence, target-language compliance, and formatting failures.
Research Assistant – Advisor: Prof. Siamac Fazli
Nazarbayev University – Water Policy Project
Nazarbayev University is a research university in Astana, Kazakhstan, with undergraduate and graduate programs and research across multiple fields.
Jan 2025 — May 2026Led computational work on water legislation across 158 countries, building multilingual processing methods and analyzing how groundwater definitions relate to governance content.
• Led the computational methodology for a 158-country study of national water legislation spanning 47 languages and 15 language families, building the multilingual OCR, legal-document processing, translation, and reproducible analysis pipeline.
• Benchmarked GPT-4.1 against Gemini 2.5 Pro on 95 translated laws using reference-free COMETkiwi and round-trip COMET; selected GPT-4.1 after it outperformed Gemini on both translation-fidelity measures.
Founding AI Engineer
Outpeer.AI
Jan 2026 — PresentImproved Gemini and Vertex AI reliability across seven services, adding validation, tracing, retries, and safer deployment and recovery processes.
• Led a reliability overhaul of Gemini/Vertex AI integration across seven FastAPI services, adding Pydantic schemas, JSON Schema validation, bounded retries, and controlled model fallback.
• Integrated Langfuse tracing and cross-service audit records for model identity, prompt hashes, validation errors, retries, fallback, token usage, and latency to support LLM debugging.
• Built asynchronous FastAPI/HTTPX services with Firestore, Cloud Tasks, bounded model-call concurrency, checkpointed batching, deadline handling, and retries.
