Taufik Adi Nugraha
Computational physicist turned data scientist
Jakarta, Indonesia · taufik.adi@openstudy.id · openstudy.id · github.com/taufik-adinugraha
Every job in this record has been the same job: build the thing that measures something nobody has measured yet, then check it against reality hard enough to find out where it breaks.
Download as PDF Same content, set for A4 — generated from this page.
Sixteen years computing the behaviour of materials from first principles — at Institut Teknologi Bandung, The University of Osaka, and for eight years at the Max-Planck-Institut für Eisenforschung GmbH in Düsseldorf. Then a deliberate change of subject: Head of Data for the IoT division of Telkom Indonesia, sensing the physical world instead of simulating it. Then an election-day quick-count platform, built alone, carrying thousands of polling stations to national media within hours of the polls closing. Now advanced analytics and AI systems at Bati Labs, and 10 public-data cases on this siteopenstudy.id.
Part of the doctoral research was published in Nature Communications — Nature Communications 11, 4127 (2020), "Exciton-driven change of phonon modes causes strong temperature dependent bandgap shift in nanoclusters" — a combined experiment-and-theory study, in which he is first-named of the two authors who carried out the DFT calculations, the first-principles half of the work. Cited 51 times. The full record, conference contributions included. The full record, conference contributions included, is at openstudy.id/cv/.
Where the time went
Bars are to scale on a single axis and nothing is tidied into a queue. The gaps are real too: the KedaiKOPI row breaks into two short segments because the work did — about two months into the February 2024 elections, and two more into the regional elections that November — and it sits inside the Telkom bar because both were freelance projects taken on alongside that job.
-
Institut Teknologi Bandung 2005 – 2009
BSc, Physics
Bandung
-
Versatile Silicon Technologies 2010
Computer Engineer
computer clustering, parallel computing
-
Institut Teknologi Bandung 2010 – 2011
Research Assistant
Computational Materials Design group · Li-polymer battery
-
The University of Osaka 2011 – 2013
MSc, Physics
Japanese Government (MEXT) scholarship via the International Physics Course · nuclear and particle physics · MSc thesis on ab initio g-tensor calculations · ISIR Nanoscience and Nanotechnology Center
-
Max-Planck-Institut für Eisenforschung GmbH 2014 – 2021
Computational Physicist
Wissenschaftlicher Mitarbeiter, Atomistic Modelling group, Düsseldorf · doctoral research on semiconductor nanoclusters, and on point defects at 3C-SiC/SiO₂ interfaces for silicon-carbide nanofibre solar cells · the affiliation on the Nature Communications paper · the institute is now Max-Planck-Institut für Nachhaltige Materialien GmbH
-
Telkom Indonesia 2022 – 2024
Data Scientist → Head of Data, IoT
January 2022 to December 2024 · Jakarta · LoRaWAN networks for asset tracking · agriculture and livestock sensors · computer vision for crowd analytics, traffic and vehicle detection, farm monitoring · NVIDIA Jetson edge inference · Maps API for commercial prospecting
-
KedaiKOPI 2024 · ~4 months of work
Quick-count platform — freelance, part-time
Built solo · two projects of about two months each, four months of work in total · pileg and pilpres on 14 February 2024, pilkada on 27 November 2024 · thousands of polling stations reporting by WhatsApp and SMS, validated on arrival, live dashboard and a feed to national media the same afternoon
-
Bati Labs 2024 – now
Advanced analytics and AI
Recommendation engine · semantic search · diagnostic engine · MCP server for context engineering, used directly by senior management · multi-agent monitoring and resolution · quantitative trading research on US equities, ML and deep learning, self-retraining, research stage · AWS, GCP, Azure
-
dakwah-lens.id 2025 – now
Community project
Non-profit media monitoring and generative AI · 27 news outlets and 4 social platforms · every citation retrieved from a verified corpus, never generated
-
openstudy.id 2026 – now
Public-data case studies
Ten public-data cases · 42 checks written before the analyses ran, 19 failed and all 19 published
One skill, four places
A CV that runs physics → telecommunications → elections → AI looks like four unrelated careers. It is not, and this is the evidence. Each row is a capability, lit in every era that used it. Read it left to right: the rows that begin in the physics column mostly do not stop there, and the one at the top — model against measurement, putting a computed answer beside a measured one and accounting for the gap — runs the entire width.
Computed EPR spectra against laboratory spectra; sensor counts against hand counts; a sample of polling stations against the official tally; and on this site, 42 checks written down before the analysis ran, of which 19 failed and all 19 are published.
Interpreting atomic microscopy images in the physics work; crowd analytics, traffic and vehicle detection, livestock monitoring, segmentation models on edge hardware at Telkom; OCR to read the surat suara — the ballot forms — during the quick count; and image embeddings in the AI systems since.
Supercomputer allocations for the physics work, with cloud services where the queue could not supply what was needed; then AWS, GCP and Azure.
Cluster building and parallel computing from 2010; then years of supercomputer workloads; now distributed pipelines.
Density functional theory and molecular dynamics for semiconductors and renewable-energy materials.
Statistical analysis and machine-learning methods from the physics work onward: forecasting, classification, imbalanced-class problems, and the evaluation that decides whether any of it is real — including deep-learning models on financial time series, in a research system that retrains as new data arrives and is judged out of sample or not at all.
Built end to end — ingestion, validation, API, interface — because the interesting problems are rarely only the model.
Maps APIs for commercial prospecting; polling stations across an archipelago; and the case maps on this site.
LoRaWAN for tracking, purpose-built agricultural and farm sensors, NVIDIA Jetson for inference without bandwidth.
Semantic search, a diagnostic engine, an MCP server for context engineering, and multi-agent environments that monitor and resolve.
Sensor and IoT streams arriving continuously at Telkom, and thousands of field reports by WhatsApp and SMS validated as they arrive on an election day that cannot be re-run.
Tap any capability for where it was used.
Sixteen years computing what could not be measured directly
Bandung · Osaka · Düsseldorf
Materials for renewable energy and semiconductors, from first principles — density functional theory and molecular dynamics, on supercomputers, at the scale of individual atoms and their electrons. The method came first: the MSc thesis at Osaka was ab initio g-tensor calculations. The doctoral work at Max Planck put that method to two problems — point defects at 3C-SiC/SiO₂ interfaces in silicon-carbide nanofibre solar cells, and why the bandgap of semiconductor nanoclusters shifts so much more sharply with temperature than a bulk semiconductor does. Both are part of the thesis; the nanocluster half is what Nature Communications published.
Underneath, they were the same exercise: compute a spectrum from theory, put it beside the one the laboratory measured, and account for the difference. That is the habit everything after this is built on. Both stints were also funded by somebody else's selection — a Japanese Government (MEXT) scholarship at Osaka, and a salaried appointment as wissenschaftlicher Mitarbeiter at Max Planck.
From simulated atoms to instrumented fields
Telkom Indonesia · Jakarta
Head of Data in the IoT division of Indonesia's national telecommunications company. LoRaWAN networks for asset tracking; sensors for agriculture and livestock; computer vision on camera feeds for crowd analytics, traffic and vehicle detection, and chicken-farm monitoring; inference pushed onto NVIDIA Jetson hardware at the edge where there is no bandwidth to send video home. Also the unglamorous, useful kind of analytics: the Google Maps API turned into a way of finding which businesses were plausible customers for any of it.
The one that could not be re-run
KedaiKOPI · freelance, alongside the Telkom role
A quick-count platform, built solo, for a Jakarta survey institute that publishes projected results to national media within hours of the polls closing. Indonesian voting runs through the morning and closes in the early afternoon, and counting at each station starts immediately — so the whole thing happens between lunchtime and the evening bulletins, in one afternoon. Thousands of polling stations — tempat pemungutan suara — reporting by WhatsApp and SMS from staff in the field; validation on arrival, because a mistyped tally is indistinguishable from a real swing until you check it; a live dashboard; and a feed to the newsrooms. Used across the legislative election, regional elections and the presidential election. No second attempt, no quiet correction the next day, and a result the whole country can check against the official count within days.
The scale of the engagement is worth stating exactly, which is why the row on the timeline above is drawn broken rather than solid: two projects of about two months each, four months of work in total and all of it inside 2024, part-time and taken on while the Telkom job was still running. That is the interesting part rather than a caveat — what it carried on those afternoons is out of all proportion to the time there was to build it.
Systems that have to be right in front of the person deciding
Bati Labs
Advanced analytics and AI to lift merchant performance: a recommendation engine, semantic search, and a diagnostic engine that pinpoints what is going wrong and says what to do about it. Internally, an MCP server for context engineering with dashboards over every process — now used directly by senior management and business staff to interrogate the company's own data by asking. And agentic environments where several agents cooperate to watch for problems and resolve them.
Alongside that, and still at the research stage: quantitative trading on the US market. Machine-learning and deep-learning models are tried against one another inside a system that runs end to end on its own and retrains as new data arrives. It is labelled research here for the same reason everything else on this site carries its status — an automated system that has not been tested out of sample, forward, on money, is a hypothesis with good tooling, and calling it anything else would be the one claim on this page that could not be checked.
The record
Separated by kind rather than pooled, because a conference contribution is not a peer-reviewed paper and a CV that blurs the two invites the doubt it is trying to settle.
Peer-reviewed journal article
F. Münzer, S. Lorenz, J. Yang, T. A. Nugraha, E. Scalise, T. Hyeon, S. Wippermann, G. Bacher
Exciton-driven change of phonon modes causes strong temperature dependent bandgap shift in nanoclusters
Nature Communications 11, 4127 (2020) · 10.1038/s41467-020-17563-0 · cited 51 times (OpenAlex, September 2026)
First-named of the two authors who carried out the DFT calculations — the first-principles half of a combined experiment-and-theory study. Part of the doctoral research.
Author contributions, verbatim: “S.W. conceived and designed the DFT calculations, T.A.N. and E.S. conducted the DFT calculations.”
Author Correction: Exciton-driven change of phonon modes causes strong temperature dependent bandgap shift in nanoclusters
Nature Communications (2025) · 10.1038/s41467-025-58988-9
First-author conference contributions · 3C-SiC/SiO₂ interfaces for silicon-carbide nanofibre solar cells
Investigation of 3C-SiC/SiO₂ interfacial point defects from ab initio g-tensor calculations and electron paramagnetic resonance measurements 2017
T. A. Nugraha, M. Rohrmüller, U. Gerstmann, S. Greulich-Weber, A. Stellhorn, J. L. Cantin, H. J. von Bardeleben, W. G. Schmidt, S. Wippermann
Ab initio electron paramagnetic resonance study of 3C-SiC/SiO₂ interfaces in SiC-nanofiber based solar cells 2016
T. A. Nugraha, U. Gerstmann, W. G. Schmidt, S. Wippermann
Understanding 3C-SiC/SiO₂ interfaces in SiC-nanofiber based solar cells from ab initio theory 2015
T. A. Nugraha, S. Wippermann
The conference items carry no DOI and no Crossref record, which is what identifies them as meeting contributions rather than proceedings papers. They are listed because they are first-author work on the SiC half of the doctoral research, not because they are journal articles.
Things that run
Three, from github.com/taufik-adinugraha.
-
openstudygithub.com/taufik-adinugraha/openstudyAstro · Python openstudy.id: ten public-data cases with their pipelines, models and evaluation. -
dakwah-lensgithub.com/taufik-adinugraha/dakwah-lensPython · TypeScript Media monitoring and generative AI for Indonesian Islamic educators, non-profit. Python API and a TypeScript front end, containerised, with a migrated relational store. -
ai-qurangithub.com/taufik-adinugraha/ai-quranPython · Django Speech recognition for Qur'an recitation, on a DeepSpeech model trained on roughly 600 hours from about 40 reciters. Begun September 2021 — a year before Whisper was released and made this class of thing a weekend project. It was not one in 2021.
What a first message should say
The useful one names the decision and the data: “we have to choose where to put X, we hold Y, and we cannot tell whether Y is good enough to decide on.” That is enough to answer yes, no, or “here is what could honestly be built from that.”
- Email · taufik.adi@openstudy.id
- GitHub · github.com/taufik-adinugraha
- Based in · Jakarta, Indonesia
The 10 cases · decisions and corrections · the commercial arrangement, stated openly
Found something wrong on this page? Report a correction — it opens a pre-filled issue — or email taufik.adi@openstudy.id. Corrections are credited by name in the errata, and one that changes a finding says so on the page.