Company: Avishkar AI (Anjaneyaai Technologies Private Limited)
Location: Remote/Mangaluru (Hybrid/In-Office)
Type: Full time Level: Senior individual contributor Reports to: Co-founder / Head of Engineering
About the role
We are looking for a machine learning engineer who has built and shipped models against live financial markets, not just backtested them in a notebook. You will own the full research to production path: sourcing and cleaning market data, engineering features that survive out of sample, training and validating predictive models, and deploying them into systems where latency, slippage and capital are real constraints.
This is a hands on role. You will write the research code, the pipelines and the production services. If you enjoy the specific discomfort of a strategy that looks brilliant on paper and mediocre in the market, and you want to understand exactly why, this role is for you.
What you will do
Research and modelling
- Build predictive models on financial time series across equities, derivatives, FX or crypto: return forecasting, volatility forecasting, regime classification, order flow prediction.
- Engineer features from raw market data including OHLCV bars, tick and quote data, limit order book snapshots, options surfaces, and where relevant alternative data such as news, filings and sentiment.
- Design and defend validation schemes appropriate to time series: walk forward analysis, purged and embargoed cross validation, combinatorial purged CV. You will be expected to explain why a standard k fold split is wrong here.
- Quantify and control for look ahead bias, survivorship bias, data snooping and multiple testing. Report deflated Sharpe ratios and probability of backtest overfitting where appropriate.
- Model transaction costs, market impact, borrow costs and slippage explicitly. A gross alpha that does not survive net of costs is not a result.
Engineering and production
- Build and maintain data pipelines that ingest, normalise and version market data at scale, with point in time correctness guaranteed.
- Own the backtesting and simulation infrastructure: event driven engines, fill models, position and risk accounting.
- Ship models to production behind APIs or streaming services, with monitoring for feature drift, prediction drift and live versus backtest performance divergence.
- Set up experiment tracking, model registries and reproducible training runs so that any result can be regenerated from a commit hash.
Collaboration
- Work directly with founders and clients to translate a trading or risk objective into a well specified modelling problem.
- Communicate results honestly, including negative results. We would rather kill a strategy early than defend it.
What we are looking for
Required
- 2+ years building ML systems, with at least 2 years applied specifically to financial markets (buy side, sell side, prop trading, market making, a fintech doing real prediction, or serious independent work you can talk through in depth).
- Strong Python: NumPy, pandas or Polars, scikit-learn, PyTorch. Clean, tested, production grade code, not research scripts.
- Deep working knowledge of time series methods: ARIMA and GARCH family models, state space models and Kalman filters, and sequence models such as LSTM, GRU, temporal convolutional networks and transformers applied to financial data.
- Practical command of gradient boosting (XGBoost, LightGBM, CatBoost).
- Solid statistics and probability: stationarity, cointegration, autocorrelation, heteroskedasticity, hypothesis testing, and the ability to reason about signal to noise ratios in the range typical of financial returns.
- Experience with market microstructure concepts: bid ask spread dynamics, order book imbalance, tick data handling, execution mechanics.
- Comfort with SQL and at least one time series or columnar store (Parquet, ClickHouse, TimescaleDB, kdb+ or similar).
- Familiarity with Docker, Git, CI, and cloud infrastructure (AWS, GCP or Azure).
Strongly preferred
- Portfolio construction and risk: mean variance and its failure modes, Black Litterman, hierarchical risk parity, factor models, position sizing and drawdown control.
- Derivatives knowledge: pricing, the Greeks, implied volatility surfaces, and modelling on options data.
- Reinforcement learning applied to execution or portfolio allocation, with a clear eyed view of where it works and where it does not.
- Low latency work: C++ or Rust, profiling, memory layout, and awareness of where microseconds actually matter versus where they do not.
- Experience with alternative data: NLP on filings, earnings calls, news feeds or social sentiment.
- Contributions to open source quant or ML libraries, published research, or strong competition results.