Building algorithmic trading infrastructure for Indian equity & derivatives markets.

Most of my day is spent on systems where correctness is not negotiable – a bug in a strategy engine is not a rendering glitch, it is a filled order. I work end to end: market-data plumbing, strategy engines, backtesting, risk gates, and the dashboards traders actually look at.

A production algo-trading and backtesting platform (orbis.vriddhix.ai), built and operated solo. Nine modules over one live market-data spine.

What it does

• Strategy Studio – visual condition-tree builder, 39 pre-built templates across technical, option-selling and price-action styles. No code required to ship a strategy.
• AlgoHub – automated strategies running as isolated OS processes, scanning 500 Nifty stocks every morning and managing positions end to end.
• Option Buyer Battlefield – weighted directional signal blending wall proximity, dealer gamma exposure, PCR trend, straddle expansion and IV percentile into one state machine.
• Gamma Blast – dealer-GEX / zero-gamma-flip read with threshold alerting.
• Smart Money · Sector Rotation · News Desk · Straddle Premium · OI Tracker – institutional flow, RRG rotation, real-time sentiment classification, and full-DTE straddle history.

How it is built

Engineering decisions I would call out

• Process-per-strategy isolation – one strategy crashing can never take down another, or the platform.
• One BaseStrategy ABC over a shared ExecutionEngine – the same strategy code runs in backtest, paper and live. No separate live path to drift out of sync.
• Paper-first by default – every strategy simulates fills through a PaperBook; live execution is admin-gated behind a risk gate and an append-only order audit log.
• Restart-proof state – per-day counters and position books are derived from the reloaded book, not in-memory, so a respawn cannot silently reset a risk limit.

Orbis is analysis and execution tooling, not advice. Not SEBI registered. The repository is private; happy to walk through architecture on a call.

algo-backtesting-platform – three self-hosted tools in one repository: Pramana, a pattern-research platform; a strategy backtester; and a live options-trading dashboard for NIFTY/SENSEX.

Pramana – प्रमाण, the valid means by which something is known – is the part I would point at. It finds chart structure in NSE equities – volatility contraction bases, breaks of market structure, fair value gaps – places each setup in market and sector context, scores it, and then records what actually happened next. Including every time it was wrong.

What it measures

Engineering decisions I would call out

• One implementation of every rule – the live scanner, the backtester and the historical X-Ray call the same function objects. There is no second implementation to drift from, and a test walks the syntax tree to enforce it rather than trusting convention.
• Nothing may see the future – a value dated t depends only on bars dated ≤ t, verified by truncating history, recomputing and demanding identical output. A look-ahead bug never crashes; it produces a backtest that looks excellent and means nothing.
• It says what it does not know – an unranked sector reads null, not 0. A win rate over zero resolved trades is null, not 0%. Absence is never quietly rendered as a bearish number, all the way through to the JSON.
• Failures are kept – failed breakouts stay in the ledger with their reason, and no view drops them by default. A hit rate computed over survivors is the single most flattering lie a research tool can tell.
• Bias travels with the numbers – a backtest that cannot resolve point-in-time index membership says so in its own results payload, not in documentation nobody reads.

Stack – Python · FastAPI · SQLAlchemy 2.0 · Alembic · pandas/numpy · SQLite → Postgres / TimescaleDB · 451 tests including no-look-ahead and engine-purity suites.

Research and execution tooling, not advice. Not SEBI registered. The trading dashboard ships in dry-run mode; Pramana has no trading path at all.

Public repositories, mostly around putting LLMs behind guardrails rather than in front of them:

• orbis-quant-agents – LangGraph agents for market analysis, with validated tool contracts and provider fallback.
• llm-support-quality-gate – quality gating for LLM-generated support responses.
• agenticworkflow-stlc – agentic workflows mapped onto the software testing lifecycle.
• agenticai-project-context – project-context tooling for agentic coding sessions.

Python · TypeScript · React · Node.js · Polars · DuckDB · Pandas · FastAPI PostgreSQL · TimescaleDB · Redis · Supabase · Docker · AWS · LangGraph

• ✉️ hello@vriddhix.ai
• ▶️ VriddhiX Wealth on YouTube – market analysis & platform walkthroughs
• 💼 Available for freelance / contract engagements

_{Most of my commit activity is in private repositories, so the graph below is fuller than the public repo list suggests.}