Tallinn, Estonia Open to opportunities
Financial research × Analytical systems × Operations
Connecting to Exponent market data.
Finance-focused EBS graduate and
Shift Operational Manager at Studioworks.
A native investment workspace connecting company research, your investment case and portfolio context. Designed and developed by me.
Native iOS · In active developmentExplore candlesticks, change the volatility in a Monte Carlo scenario and see a valuation respond to a new discount rate. Recorded English app footage, with historical market data and illustrative model results. Exponent is a native iOS app in active development.
Music: Titan — Scott Buckley, released under CC BY 4.0. Edited for this film.
Choose a view. Inspect the actual app.
Explore companies, inspect the fundamentals and follow the sources. Turn an interesting business into a question worth investigating.
Research lenses, valuation, investment cases, scenarios and deliberate review.
From a research question to a working system.
Explore the evidence behind each project.
How can evidence stay connected to a financial decision?
A native workspace connecting source evidence, financial models and portfolio risk.
Built a research workspace that keeps source states, models and risk checks alongside one another.
Product architecture · interface · native applicationLead shift-level operations across two portfolios. Connect workforce planning, service availability and people development with the systems that make the work measurable.
I tested whether search activity, news and market sentiment improve stock-return models. The answer depends on the question.
Machine learning added explanatory power across the full panel.
The European gain alone was not statistically significant. The regional result matters as much as the headline.
Change the volatility. Extend the horizon. See how the range of possible outcomes grows, even with the same assumed return.
All outcomes follow the same lognormal model. Only the selected assumptions change.
Research, software and operations are connected parts of how I work.
Explore skills & evidenceFrame a testable question. Compare models. Make the limits visible.
Python · SQL · BigQueryTurn recurring work into interfaces, pipelines and useful tools.
TypeScript · React · Rust / TauriConnect capacity, service performance and the people behind it.
Planning · Reporting · LeadershipCareer, education & projects
A working life and a finance degree developed together. Trace the path from the service floor to operational leadership, research and software.
Joined a 24/7 service floor at Studioworks OÜ — service delivery inside the cycle, quality standards at the point of contact.
Took on escalation and incident handling, shift coordination and quality evaluation, expanding from individual service delivery to a wider operational role.
Finance-focused degree begins at Estonian Business School, taken alongside the full-time operational role.
12-week field sales programme on an independent territory, run between operational seasons.
Operations, workforce planning & analytics: capacity plan ownership, the availability KPI (99.84% across 49,900 agreed service hours), the monthly reporting model, hiring decisions.
Degree completed with honours — three years of evening study carried next to a 24/7 operation.
25,876 firm-week observations across 99 U.S. and European firms. XGBoost 0.2146 same-week out-of-sample R² against a 0.1624 linear benchmark; the European gain is not significant and the one-week-ahead null is stated.
A native decision workspace: 96 registered views, 67 native commands, six evidence layers. Research and paper state only — documented in the thesis, no live trading claimed.
Company-wide internal product with 136 personal dashboards, role-based access and HR integrations. Employees receive feedback directly; an automated weekly digest gives leadership a shared view. Scoring consistency and coverage follow-through remain management responsibilities.
Daily operational performance evaluation tool in production: screenshot evidence to auditable monthly workbooks, with a two-phase human review gate.
Four instruments computing in the page — simulation, frontier, stress and derivatives — with no external numerical dependencies and the closed-form answer kept on screen as the check.
Each milestone is part of the same practice:
understand the problem, build the system, check the result.
Case studies
Five applications of the same discipline: turn source data into research, reporting or decisions people can inspect. Explore the result, how it works, and the limits of each system.
Better same-week fit in the full panel; no robust next-week predictability
Boundary: At a one-week-ahead horizon, no model produced robust out-of-sample predictability. Reported explicitly; not presented as a trading rule.
Brings source evidence, model results and risk checks into one research workspace
Boundary: Research and paper state only. No live trading performance is claimed and no production execution-control readiness is claimed.
Turns evaluation screenshots into reviewed, traceable monthly reporting
Boundary: The interface shown on this site is an anonymised reconstruction. It does not claim perfect extraction or automatic accuracy.
Lets visitors test portfolio assumptions and compare numerical results with analytical checks
Boundary: Educational instruments on user inputs. No market data, no calibration to real assets, and nothing on the screen is investment advice.
Replaces the manual relay with direct access to evaluations and a weekly leadership digest
Boundary: The platform surfaces coverage gaps; closing them remains a management routine. Scoring consistency is governed by the written framework and manager cross-checks — not by the tool itself.
Research Published study, reproducible
Better same-week fit in the full panel; no robust next-week predictability
At a one-week-ahead horizon, no model produced robust out-of-sample predictability. Reported explicitly; not presented as a trading rule.
Yahoo Finance, Finnhub, Google Trends, NewsAPI and FinBERT / Loughran–McDonald scoring. ≈130,000 outbound calls.
| Project | Bachelor's thesis | NEXUS Terminal |
|---|---|---|
| State | ResearchPublished study, reproducible | Paper stateInstalled build, no live execution |
| What it delivers | Better same-week fit in the full panel; no robust next-week predictability | Brings source evidence, model results and risk checks into one research workspace |
| Scale & evidence | 0.2146same-week OOS R² | 96registered views |
| Stack | PythonSQLBigQuery | TypeScriptRustTauriReact |
| Scope & limits | At a one-week-ahead horizon, no model produced robust out-of-sample predictability. Reported explicitly; not presented as a trading rule. | Research and paper state only. No live trading performance is claimed and no production execution-control readiness is claimed. |
Research · EBS bachelor’s thesis, 2026
Linear and machine-learning evidence from the U.S. and European markets. Estonian Business School bachelor’s thesis, 2026 — an auditable pipeline that turns raw price, attention and news data into interpretable models, then validates them under a strict chronological design.
Test whether abnormal search activity, news intensity and financial-news sentiment add incremental information to short-horizon equity-return models after controlling for market, volatility, liquidity, size, value, momentum and sector effects.
Chronological governance
The panel · figure
The 2025 block is touched once, after the model is frozen. Nothing to its left is re-fitted afterwards—which is what makes the out-of-sample figure below an out-of-sample figure.
Out-of-sample evidence · frozen 2025 test window
Identical held-out observations. Same-week horizon (h = 0).
Diebold–Mariano 7.13, p < 0.001 — the gain survives the test.
This measures same-week explanatory power. It is not a next-week return forecast.| Market | Pooled OLS | XGBoost | Interpretation |
|---|---|---|---|
| Full panel | 0.1624 | 0.2146 | Diebold–Mariano 7.13, p < 0.001 — the gain survives the test |
| United States | 0.1740 | 0.2237 | Strongest regional result across the study |
| Europe | 0.1479 | 0.1489 | Incremental machine-learning gain not statistically significant |
ASVI was the highest-ranked information-flow feature in SHAP attribution.
Thesis · Figure ML-A · SHAP attribution
A pooled linear model has one ASVI coefficient and must pick a single sign. Attention behaved differently in a rally than in a panic-search week—and reproducing that is where the machine-learning gain in the table above actually comes from.
Forecasting models did not outperform the mean-return benchmark. The thesis states this explicitly and does not present the result as a standalone trading rule. Same-week explanatory power and next-week forecasting power are different claims, and only one of them survived the test.
End-to-end capability
Model-risk discipline
Every predictor is available by the close of week t. Nothing in the feature set could only be known later.
Scaling and encoding are fitted on training data only, then frozen for validation and test.
A 243-setting hyperparameter grid is selected on 2024 validation alone; the 2025 window is touched once.
Insignificant and regime-sensitive outcomes stay in the thesis instead of being trimmed out of the story.
Data sources
Yahoo FinanceFinnhubGoogle TrendsNewsAPIFinBERTLoughran–McDonaldProfessional record
Five years at Studioworks, progressing from service delivery to shift-level planning, reporting and operational analysis. This record connects my responsibilities with measured outcomes.
Studioworks OÜ · Tallinn, Estonia · 2021–present
Workforce planning, reporting & operational analytics
Current role · since 2025Jan–May 2026 planning cycles. Downtime is attributed by driver rather than absorbed into a single figure.
19 service lines · two portfolios
Plan vs actual · KPIs owned and reported
Contractual KPI met
Downtime attributed by driver, not absorbed into one figure
Systematic drift — not one-off variance
Re-based the service-hour requirement; flagged capacity risk
Headcount overstated available hours
Coverage adjusted per service line before it reached service
The data existed; the KPI did not use it
KPI rebuilt formula-driven; auditable to source
Built in-house · in production use
Advanced Excel · two portfolios · up to 19 service lines
Eight numbered steps from total possible hours through availability, FTE and coverage per line, to error rates and mean resolution time.
Reconciled monthly against five source feeds and reported plan-vs-actual with variance commentary.
Python · SQL · web application · 136 employees
Self-initiated. Evaluations, errors, attendance, attitude and skills in one place, with role-based access and database-backed logging.
Replaced scattered manual records with a single auditable source for absence, error and skill data.
HTML / JS · Excel
The framework the operation is scored against, plus an offline scenario trainer and a monthly schedule-review matrix.
Standardised how quality and coverage data enter the reporting cycle.
Additional experience
Ran an independent sales territory through a 12-week field programme, planning daily activity from conversion data and adapting targets weekly in a fully autonomous international environment.
Skills & tools
Financial risk, quantitative methods, engineering and operations. Explore the skills I use and the projects where each has been applied.
Filter by a body of work, then follow the evidence links to see where each skill has been applied.
Delivery & leadership
Primary evidence
Research figures, product screens and system architecture. Open any item to inspect the original at full resolution.
Open artefact Thesis · Figure 7 · p.48
Open artefact NEXUS · Installed build
Open artefact NEXUS · Research build
Open artefact Thesis appendix · Figure A1
Academic distinction
Cum laude
Finance-focused degree culminating in an empirical thesis that connects market attention, machine learning and an implemented analytical system.
Additional risk foundation · Disney, $100k position
Reads the loss straight off the observed distribution — the 5th and 1st percentile of what actually happened.
Recognition
Winner — Baltic Essay Contest in GermanAwarded for written argument in German, alongside a 12-week autonomous sales programme run entirely in English in Texas.Risk console
The panels carry the same verified figures published across this site; the VaR and valuation calculators run right here in the page on the formulas shown. No market feed, and nothing represents a live position.
Linear benchmark vs XGBoost · axis to 0.25
Daily · scales the published Disney study to your position · not advice
Two-stage DCF · 5y explicit + Gordon terminal · your inputs only — not advice
ASVI flips sign across four 2025 regimes
NEXUS · portfolio edition — published figures, plus calculators that run right here.
Type `help`, or select a module on the left.
Quant Lab
Explore simulation, portfolio allocation, stress and options. Change the inputs and compare the results with analytical checks. Educational models, with assumptions visible and no live market feed.
Instrument I · simulation
Geometric Brownian motion, monthly steps, seeded — the same inputs always reproduce the same run.
| terminal (10y) | simulated | closed form (GBM) |
|---|---|---|
| median wealth | — | $162,824 |
| 5th percentile | — | $74,623 |
| 95th percentile | — | $355,275 |
| P(end below start) | — | 15.2% |
| median max drawdown | — | path property — no closed form |
The simulated and closed-form columns must agree — that agreement is the check that the engine is right, and the test suite enforces it. Your inputs · standard model · no market data · not advice.
Instrument II · optimisation
4,000 long-only random portfolios against the closed-form Markowitz frontier. The frontier itself is unconstrained, so it may short.
| frontier @ μ* | weight |
|---|---|
| Asset A · growth | 30.7% |
| Asset B · income | 24.6% |
| Asset C · diversifier | 44.7% |
| portfolio σ | 8.7% |
Closed-form Markowitz on your parameters. Negative weights are short positions — the unconstrained frontier allows them and they are shown, not hidden. No market data · not advice.
Instrument III · risk
The same three assets under a named shock. The waterfall shows which assumption change does the damage — usually correlations.
| daily · 99% | baseline | Risk-off |
|---|---|---|
| Value at Risk | $4,192 | $9,411 · ×2.2 |
| Expected shortfall | $4,812 | $10,787 |
| portfolio μ · σ (annual) | 6.5% · 11.6% | |
Parametric under normality — a deliberate simplification, and the reason the thesis reports historical VaR beside it. Your inputs · no market data · not advice.
Instrument IV · derivatives
Black–Scholes in closed form on the left; on the right, what a desk actually lives with — the P&L of a delta-hedged short call under discrete rebalancing.
| quote | call | put |
|---|---|---|
| price | 9.41 | 6.46 |
| delta Δ | 0.599 | -0.401 |
| gamma Γ | 0.0193 | |
| vega / vol pt | 0.387 | |
| theta / day | -0.015 | |
A perfectly hedged book would sit at zero. What remains is discretisation risk — tighten the rebalance and σ shrinks roughly with √frequency; the test suite asserts the ordering. Hedged at the pricing vol · no market data · not advice.
Financial research. Analytical systems. Operations.
Let’s find the question worth working on.
A role where research, systems thinking and operational experience can make a difference.
“For to be possessed of a vigorous mind is not enough; the prime requisite is rightly to apply it.”
A native investment workspace connecting company research, your investment case and portfolio context. Designed and developed by me.
Native iOS · In active developmentExplore candlesticks, change the volatility in a Monte Carlo scenario and see a valuation respond to a new discount rate. Recorded English app footage, with historical market data and illustrative model results. Exponent is a native iOS app in active development.
Music: Titan — Scott Buckley, released under CC BY 4.0. Edited for this film.
Follow the story. The interface stays with you.
A price chart tells you what moved. A report tells you what happened. Neither remembers why you invested. The reasoning gets lost between tools.
Six connected capabilities.
Explore how the pieces work together.
Explore companies through cash-flow quality, valuation and the direct sector weights of your portfolio. Each candidate carries the metric, report date, source and questions still worth asking.
A defined research scope, not an all-market ranking or a buy signal.
The app supports research, comparison and an explicit decision process. Confirming a plan is separate from executing a trade; portfolio reconciliation remains deliberate.
Exponent is an independent application in active development. These are actual app screens with demonstration data. This showcase is not a public download or a brokerage service.
Let’s talk about Exponent
