Projects

A focused set of systems across backtesting, perp execution research, funding research tooling, and onchain games.

Watson

Backtesting engine for TradingView power users

Research system

Backtesting engine for people who already live in TradingView and need to stop eyeballing charts. It turns exported signals and market data into replayable tests, candidate strategies, walk-forward validation, and guardrails against fake edge.

TradingViewBacktestingValidationTradingData
Read proof notes →

Rialto

AI-assisted execution and research system

Independent research system

Rust/Python system targeting Hyperliquid execution and reproducible research. I own the architecture, data-source research, product UX, requirements, acceptance criteria, testing, and review loops; three reviewed contributions reached main while Hyperliquid parity remains under review.

RustPythonHyperliquidReplayRisk Engine
Read case study →

vc-analyzer

Graph/provenance workflows for funding research

11K+ funding rounds indexed

Research system built on 11,000+ indexed public crypto funding rounds and their investor relationships. Maps investors, co-investment edges, provenance, and research workflows so VC diligence can move from scattered spreadsheets to repeatable graph-backed analysis.

Graph DBPythonTypeScriptDataVC
Read case study →

Warpacks

Onchain collectible pack experience on Starknet

Grant funded

Founder-led onchain game/product work: game mechanics, economy design, and Cairo smart contracts for a collectible pack experience on Starknet.

GamingStarknetCairoGame Design
GitHub →

Vara Arena

Onchain PvP game on Vara

Grant funded

Founder-led PvP game built around onchain mechanics and Rust smart contracts on Vara/Gear, from game design through implementation and grants.

GamingVaraRustPvP
GitHub →

Case studies

Proof Notes

Watson

Research layer for Rialto: dataset alignment, feature work, replay, and validation

101,311 Bybit rows

Problem

Trading research is easy to fake by accident: misaligned datasets, future leakage, cherry-picked cohorts, and candidates that look strong only because the validation loop is weak. Watson exists to make Rialto research harder to fool.

Approach

Built a research layer around canonical dataset alignment: 101,311 Bybit 1m rows, 3,377 CSV overlay rows, and 286 reconstructed events. On top: feature registry, cohort mining, candidate generation, walk-forward validation, Time Machine replay, no-leakage checks, and promotion guardrails.

Why it matters

The system changes the question from “does this signal look good?” to “does this signal survive clean data, realistic replay, and promotion criteria?” That is the difference between a demo and decision infrastructure.

Case Study

Rialto

AI-assisted execution and research system with deterministic Rust risk controls

Independent research system

Problem

Manual crypto trading loses money because humans react late and trade on emotion. I wanted to test whether an LLM reasoning layer inside a disciplined system could make better decisions — not as a black box, but as one component with clear boundaries.

Approach

TradingView alerts flow into market/context analysis, LLM signal evaluation, and a Rust risk engine that owns sizing, exposure, execution, and guardrails. Watson strengthens the research side by testing which candidates deserve promotion before they can be considered for live execution.

Key decision

Separate reasoning from execution. The model can interpret noisy signals; deterministic software enforces risk. One component should not do both.

Case Study

vc-analyzer

Graph/provenance workflows for crypto funding research

11K+ funding rounds indexed

Problem

Evaluating a crypto deal means understanding investors, co-investors, provenance, outcomes, and market context. A spreadsheet stores the facts, but it does not make the network or source trail easy to interrogate.

Approach

Indexed 11K+ public crypto funding rounds and their investor relationships into graph/provenance-oriented workflows so research could move through investors, co-investment clusters, portfolio companies, and supporting sources without rebuilding the context manually each time.

Outcome

Used for DD and deal-flow triage: faster first-pass research, clearer investor structure, and less time wasted stitching together source material before the real judgment work begins.

Let's talk

If you're hiring for a product role where technical depth, user understanding, and delivery all matter, book 30 minutes and I'll come prepared.

Product LeadArmenia · Remote · UTC+4
Book a 30-min call
Telegram X GitHub LinkedIn