Acciolytix — AI product & engineering intern, Acciolytix

• Product engineering, PostgreSQL/SQL, XML ingestion and data validation for an enterprise AI workspace.
• Built and debugged production-facing workflows in Next.js, React and TypeScript; audited generated SQL for Text-to-SQL features.

Digitalization internship, Astana City — Digitalization intern (paid), Department of Architecture, Urban Planning and Land Relations of Astana

• Digitized, classified and quality-checked official architecture and land records.
• Worked inside municipal archival systems, document retrieval and administrative workflows.

NanoTranslate-S3 — Independent researcher

• Co-designed a 999,040-parameter RU→EN encoder–decoder Transformer, a lossless byte-piece tokenizer, row-wise INT8 weights and a C++ runtime for the ESP32-S3.
• Verified the PyTorch → INT8 → C++ path end to end; the preprint leaves board latency and held-out BLEU blank until measured.

Sycophancy as a pathogen — Independent researcher, ThinkLab International Research Competition 2026

• Designed a multi-turn sycophancy evaluation across GPT-5.4, Claude Sonnet 4.6 and Gemini 3 Pro with response-level coding and a Drift Score.
• Fine-tuned Qwen2.5-0.5B-Instruct with DPO (LoRA, 4-bit NF4) on 44 preference pairs; observed drift fell from 33.3% to 12.5% in a single pilot run.

GraphTrust — Research engineer, team Cipher (with Shyngyskhan Sagyt and Arpith Sivaramakrishnan), NSRI Summer Research Hackathon 2026

• Co-developed a typed graph system for multi-hop identity risk, bounded path search, analyst-budget ranking and workflow-preserving counterfactual remediation.
• 225/225 verified runs; NDCG@10 of 0.125 against 0.002 for untyped traversal, with equal risky-identity recall (0.758).

Corrosion prediction across publications — Co-author, with Ramazan Zhumash (grade 11); supervisor Nurbol Kelgenbaev, NIS Astana-Nura · Chemistry section

• Co-authored a school research project (with R. Zhumash; supervisor N. Kelgenbaev) on transferring ML predictions of pitting potential across publications, with expected-error estimation and a pre-specified external test.
• Internally, rejecting high-risk predictions cut MAE from 0.249 to 0.171 V at 61.8% coverage; on 40 external measurements the rule failed (0.281 → 0.324 V) and is reported as such.

ArtEco — Independent project

• Built a 20×20 multi-agent economy in pure NumPy with learned barter and a 4-bit message channel; ran six experiment families.

Deep-SAR Oil Spill Segmentation (Refined) — Curator

• Published a refined version of the Deep-SAR Oil Spill dataset on Kaggle: corrected 38% of training and 50% of validation segmentation masks (CC BY 4.0). Used in a peer-reviewed paper in Discover Computing (Springer Nature, 2026).

Kazakh Classical Poetry — Abai & Makataev — Curator

• Published a cleaned corpus of 233 Kazakh poems by Abai Kunanbaiuly and Mukagali Makataev for NLP (CC0).

TRACE — Founder & solo developer, AEROO International Space AI Competition

• Built a maritime monitoring stack: YOLOv8-OBB vessel detection, U-Net/ResNet-34 oil-spill segmentation on SAR, XGBoost over 7.17M AIS records, FastAPI/Leaflet interface and generated reports.

SunCast — Researcher, with Dr. Dana Alina (PhD in Astrophysics, Nazarbayev University)

• Built SunCast, a solar-flare forecasting system integrating NASA and NOAA satellite data with multimodal space-weather signals; developed and evaluated RNN, LSTM, Transformer and ensemble pipelines in PyTorch, reaching 97.07% accuracy and 95.26% recall, with Dr. Dana Alina (PhD in Astrophysics, Nazarbayev University).
• A later pre-registered audit (Oct 2026) found no confirmed 5% Brier improvement on 2026 data and is reported as a negative result.

EpiCare — Independent project

• Built an ESP32-S3 wearable with inertial, ECG and SpO₂ sensing; converted an XGBoost model to C with m2cgen for on-device inference.
• 97.0% accuracy and 98.9% recall on a 901-window fall-vs-activity split; added pre-alert cancellation and Wi-Fi telemetry.

Mitra — Builder

• SAT practice platform (adaptive modules, spaced repetition, vocabulary trainer) merged with an opportunity aggregator: Python collectors, LLM extraction, pgvector, Telegram moderation.

Mitair — Tech Vision hackathon

• Android app (Expo/React Native, FastAPI, Supabase) comparing departure times and routes by expected PM2.5 exposure; LightGBM forecast beat persistence by 21.7% at 6 hours on a temporal holdout.

MedArchive — MedPartners hackathon case

• Built a searchable medical price-list archive from 10 source archives and 8,877 structured price rows; shipped a prototype with API and Swagger documentation.

JerCase AI — Team baash, one of four, GovTech Camp 2026 (selection stage)

• Linking land-case documents to cadastral parcels with OCR, multilingual embeddings and a CatBoost YetiRank reranker over PostGIS.

1st place — AEROO Space AI Competition

• I degree diploma, Space AI section, International Science Competition in Space Research “Exploring the World of Science” (Astana, April 2026). Project: TRACE.

Gold Honour — International Astronomy & Astrophysics Competition

• Top 2% of all participants; 19 points in the 20-question supervised final.

Grand finalist · Top 25 — NSRI Summer Research Hackathon

• Grand finalist (Top 25 of 100 finalist teams), team Cipher. Project: GraphTrust.

75th of 4,370 — Kaggle: Predicting Heart Disease

• 75th of 4,370 teams (top 1.7%), Kaggle Playground Series S6E2 — tabular binary classification.

1st place — Science Fair at NIS Astana-Nura

• 1st place, Technology section, IV regional conference of student science projects.

Top 4 — Decentrathon 5.0

• Top 4 of 5 finalist teams, “Legislative Entropy” case, AI inDrive track, team baash (Pulse 109: routing, retrieval and analytics for citizen appeals).

1st place — NIS Hackathon 2026

• Awarded for innovation, problem-solving and teamwork.

8th of ~700 teams — Halyk Bank Agentic Challenge

• Project: auditable covenant decision engine.

1st place ×2 — School olympiads

• 1st place, school astronomy olympiad; 1st place, world history (school stage of the national olympiad), 2024–25.

2nd of 11 · 4th of 45 — Kaggle community competitions

• 2nd of 11 (Image Classification Challenge) and 4th of 45 (Cyprus AI Camp: Binary Classification in Space), Kaggle community competitions.

Qualified for the pre-final — International Youth Math Challenge 2026

• Passed the qualification round; qualified for the pre-final (October 2026).

Mostly Matter — Founder & author

• Science short-video channel: 1.1M+ cumulative views across 26 videos, 4,200+ followers (October 2026).

Quasar Astronomy & Astrophysics Club — Co-founder & deputy president

• Co-founder and deputy president; organize lectures, research sessions and events with PhD scientists from Nazarbayev University.

Chess — Player & teacher

• First-category player (2022 certificate); about 190–200 hours of teaching younger students; students reached third-category ratings.

Volunteering — Volunteer

• Event volunteer — NIS Run Fest 2024 (certificate of appreciation), KazRockets, Barys Arena.

Junior Academy, New York Academy of Sciences — Research participant, The New York Academy of Sciences

• Active 2025–26 member; team research project on solar energy.

School physics team — Team member, grades 7–8, NIS Astana-Nura

• Member of the school physics olympiad team in grades 7 and 8.

AEROO Space Settlement Competition 2025 — Team Perseus, one of four authors, AEROO

• Team Perseus: design proposal for a 100,000-person rotating habitat in the Kuiper Belt (junior category).

Hugging Face LLM Course — Hugging Face

• Certificate, chapter 3 “Fine-tuning Language Models” (March 2026).