Senior Machine Learning Engineer

Novaex.ai; Bengaluru; On-site; INR 1,200,000 - 1,800,000; Full time

Job summary

Novaex.ai is hiring a Machine Learning Engineer to build the predictive brain of our platform. You will own feature pipelines, model deployment, and end-to-end ML operations in a production environment focused on real-time trading signals.

You will work closely with the COO and CTO, bridging quantitative research and production engineering, and contribute to backtesting, feature catalogs, and model governance for robust decision making.

Qualifications

• 4–6 years hands-on ML engineering, with production work on time-series or financial data.
• Experience building feature stores and backtesting harnesses for live signals.
• Ability to deploy multi‑modal models and reference real-time trading signals.

Responsibilities

• Build bronze→silver→gold feature pipelines that are versioned and testable.
• Design and deploy time-series models to forecast price, volume and volatility.
• Develop robust NLP pipelines for sentiment, events and entity extraction.
• Fuse signals from NLP, market microstructure and technical features to generate signals.
• Lead backtesting with walk-forward validation, latency accounting and cost modeling.
• Own end‑to‑end ML operations: data ingestion, training, deployment, and monitoring.

Skills

Deep learning

Feature engineering

Quant/data fluency

NLP expertise

Statistical rigor

Engineering discipline

Communication

Shipping track

Curiosity

Tools

PyTorch

TensorFlow

Pandas

SQL

Airflow

MLOps

Job description

Location: In-office, Koramangala, Bangalore Experience: 4–6 years of hands‑on ML engineering, including production work on time‑series or financial data Reports to: COO, working closely with the CTO

The Mission

The global commodities market operates in a storm of noise. Prices are driven not just by supply and demand, but by a complex interplay of geopolitical events, breaking news, and market microstructure. At Novaex.ai, we are building the definitive AI‑powered operating system to decode this complexity, starting with base metals (copper, aluminium, zinc, nickel, lead and tin across the LME, COMEX, SHFE and MCX). We don't just want to display data; we want to predict its movement.

We are looking for an exceptional Machine Learning Engineer to join our core team. This is not a role for someone who wants to tweak hyperparameters on static datasets. This is a role for a builder who wants to deploy multi‑modal models that ingest real‑time chaos – price ticks, news sentiment and volume flows – and output clear, actionable trading signals. You will help build the predictive brain of our platform, and the systematic foundations it runs on.

Your Mandate

Reporting to the COO and working hand in hand with the CTO, you will bridge the gap between quantitative research and production engineering. You will own the full lifecycle of our predictive models, from hypothesis and feature design to live signal generation.

We already have raw and partially cleaned market data flowing in. The first priority is the layer that turns it into a signal: a clean, tested, versioned feature platform and a backtesting harness we can trust. Everything our ML roadmap builds sits on top of it. If you would rather jump straight to training a Transformer than first make sure the features feeding it are correct, point‑in‑time safe and reproducible, this is not the right fit.

Your core responsibilities will include:

• Building the Feature Platform: Design and build our bronze → silver → gold feature pipelines across price, volume, inventory, spreads, macro and text data. Features must be config‑driven (no hardcoded symbols, dates, thresholds or paths), point‑in‑time correct, unit‑tested, versioned, backfillable, and documented in a feature catalog anyone on the team can read.
• Building the Predictive Core: Design and deploy time‑series models to forecast price, volume, volatility and spreads, working directly with tick‑level and intraday data to capture market dynamics. Start with honest baselines (naive, statistical, gradient boosting) and use state‑of‑the‑art deep learning (LSTMs, GRUs, CNNs, Transformers) where it measurably earns its place.
• Decoding Market Sentiment: Construct robust NLP pipelines that perform sentiment classification, event detection and entity extraction on diverse unstructured sources – exchange notices, corporate announcements, news, social media and analyst commentary.
• Multi‑Modal Fusion: Combine hard numbers with soft signals. Develop models that fuse sentiment scores with market microstructure (order books, quotes) and technical features (OHLCV) to generate superior trading signals.
• Rigorous Backtesting & Validation: Lead a walk‑forward backtesting framework that prevents look‑ahead bias and leakage, accounts for realistic costs and latency, and stress‑tests models across market regimes to ensure robustness and minimise overfitting. You decide when a model is ready, and you can defend that decision with evidence.
• End‑to‑End ML Operations: Own the code that runs the models: data ingestion, feature engineering, training and deployment, plus the dashboards and alerting that track model drift, latency and performance in real time.
• Visibility & Communication: Write short design docs before building anything significant, share a written weekly update (shipped, learned, blocked, next), and demo your work regularly. Anyone on the team should be able to see what you are building and why.

What Success Looks Like in Your First 90 Days

• Day 30: You have audited our existing data pipelines and delivered a written assessment plus a feature platform design doc, agreed with the COO and CTO.
• Day 60: A production gold feature layer for our core metals is live, with data‑quality checks, tests, lineage and a documented feature catalog. The walk‑forward backtesting harness is running.
• Day 90: At least one forecasting model beats naive baselines out‑of‑sample, is running live in shadow mode with monitoring, and you can explain every step of its logic to both an engineer and a metals trader.

What You Bring

We are seeking a high‑potential engineer who combines strong theoretical knowledge with the ability to ship production‑grade code. The ideal profile includes:

• Deep Learning Proficiency: 4–6 years of hands‑on experience with Python and frameworks like PyTorch or TensorFlow, specifically applied to time‑series forecasting and sequence modelling. You also know when a simpler model is the better choice, and can prove it.
• Production Feature Engineering: You have built feature pipelines that other people and models depend on. You know point‑in‑time joins, as‑of logic, survivorship and look‑ahead bias, and how to backfill without contaminating history. Experience with a feature store (Feast, Tecton or a well‑built in‑house equivalent) is a plus.
• Quant & Data Fluency: Comfortable with financial datasets (OHLCV, tick data, order books, futures curves, rolls and spreads). You understand market microstructure, volatility surfaces and derived indicators.
• NLP Expertise: Experience with modern NLP techniques (Transformers, HuggingFace, BERT/RoBERTa, FinBERT) for sentiment analysis and text classification.
• Statistical Rigor: You don't just look at accuracy; you understand precision and recall, calibration, Sharpe ratios and drawdowns, regime stability, and the dangers of look‑ahead bias. You know how to validate a model properly and can spot flawed logic, including your own, before it reaches production.
• Engineering Discipline: Strong Python and SQL; modular, typed, tested code with CI and code review; config over hardcoding. Experience with Pandas and NumPy, an orchestrator (Airflow, Dagster or Prefect), data‑quality tooling (Great Expectations, pandera or similar), and MLOps best practices (model versioning, experiment tracking, monitoring) is essential.
• Clear Communication: You explain your reasoning simply, in writing and in person, and raise problems early with a proposed fix.
• A Track Record of Shipping: You can point to systems you built end to end that ran in production, and walk us through one in detail.
• Curiosity: You are fascinated by the markets. You want to understand why a price moved, and how to teach a machine to see it coming.

Nice to Have

• Experience with commodities, especially base metals, or with exchange data from the LME, COMEX, SHFE or MCX.
• Cloud deployment experience and columnar data at scale (Parquet, Polars, Spark or DuckDB).

How We Work

• In office, together. We are an early team, and we build fastest side by side. Around major product milestones we occasionally put in extra days as a team, and we value people who lean in when it matters.
• If it isn't in the repo and documented, it doesn't exist. All code is version‑controlled and reviewed; all features and models are documented.
• Write it down. Design docs before big builds, weekly written updates, and decisions recorded where the team can find them.
• Ownership. Deadlines slip sometimes. When they do, we flag it early, explain why, and come with a plan.

Our Hiring Process

• Round 1 – COO (45 minutes): A semi‑technical conversation about how you think, how you break down problems, and how you make decisions.
• Round 2 – CTO: A fully technical deep‑dive into ML, data engineering and your past work.
• Round 3 – In office: A technical problem to work through with the team, combined with a culture‑fit conversation.

The Opportunity

This is a rare opportunity to apply advanced ML to a high‑stakes, real‑world domain, and to build the ML foundation of a fast‑moving startup from the ground up. We offer a competitive compensation package including a strong salary and equity. You will work alongside industry veterans and brilliant engineers, building a product where your algorithms directly impact strategic decision‑making. At Novaex, you aren't just training models; you are engineering the future of commodity intelligence.