Data & Machine Learning Engineer

Role: Data & Machine Learning Engineer

AI-Native Product Engineering | WFO Bangalore | [CTC band] | 4+ years

About TPH

The Product Highway is an AI-native product strategy and engineering firm that partners with businesses from the earliest spark of an idea all the way through to enterprise scale. We grew from zero to multi-million ARR within our first year, working with clients across India, APAC, Europe, and North America.

The conventional product/software industry is coming to an end. The line between product development and business development is disappearing. The best talent won't spend a career maintaining one product. They'll operate as forward-deployed product managers and engineers, building businesses end to end in small, high-leverage teams, one after another. That's the vision we're building toward.

AI got very good at the how: shipping code faster than ever. But almost nothing has changed in the what: deciding what to build, why, and in what order. That's where we live. We believe software is an approximation of the real world, and what matters isn't lines of code or sprint velocity. It's whether the solution actually maps to how a business works, how customers think, and how value gets created.

From in-house AI project managers to near-universal adoption of tools like Cursor and Claude, we've rethought every conventional process from first principles. AI isn't a feature we offer. It's how we think, build, and deliver. The result: legacy systems rebuilt in under 6 months, enterprise platforms built in 3, mobile applications shipped in under 1.

We only take on problems we find genuinely interesting and worthy of being solved, and we refuse to ship anything we wouldn't stand behind. Every hire gets us closer to that standard.

Who we're looking for

Regardless of role, every person at TPH shares these traits:

• End-to-end owners. You own outcomes, not tasks. If something you're responsible for falls through a crack, that's on you. Not because someone assigned it, but because you wouldn't have it any other way.
• Clear, direct communicators. Bad news doesn't get better with age. Context doesn't transfer through vague Slack messages. You surface problems early and communicate with precision.
• Uncompromising on quality. You have a quality bar that's yours, not your manager's. You won't ship something you wouldn't stand behind, even under deadline pressure.
• AI-obsessed. You see AI as how work gets done, not a nice-to-have. If you're still doing something manually that AI could handle, you feel that as friction, not normalcy.
• Structured thinkers. When faced with an ambiguous problem, you break it down, reason through the trade-offs, and arrive at a position. You don't wait for someone to tell you the answer.
• Experienced enough to use AI systematically. You have enough depth in your craft that AI makes you dangerous, not dependent. You direct it, evaluate its output, and know when it's wrong.

Why we're hiring

The AI systems we ship are only as good as the data underneath them: clean pipelines, well-engineered features, labelled datasets that hold up under scrutiny, and evaluation sets that actually mean something. That layer is now the bottleneck on most of what we build - recommender systems, moderation pipelines, retrieval-heavy products, and vision and vision-language workflows. We need an engineer who owns it end to end: someone who has moved data from raw to production-ready, worked shoulder to shoulder with annotation teams, and knows the difference between a notebook that runs once and a pipeline that survives real volume.

The work

You'll build the data and model layer across TPH's client portfolio: ingestion and transformation pipelines, feature engineering on unstructured text, annotation workflows that feed training and evaluation, recommender systems for e-commerce and content surfaces, and inference pipelines for vision and vision-language models. The projects change. The bar doesn't: production quality, measured behaviour, no vibes-based shipping.

Stack exposure: Python, SQL, Postgres, pandas and Polars, pipeline orchestration (Airflow, Dagster, dbt), PyTorch, embeddings and similarity search (pgvector, FAISS), recommender and collaborative-filtering libraries, vision and vision-language models (Grounding DINO, CLIP-family VLMs), experiment tracking and eval harnesses, AWS.

What you'll achieve

• Own the data layer behind AI systems real businesses depend on, across multiple domains, not one product for years
• Build genuine depth in the parts of ML engineering that matter in production: data quality, feature design, evaluation, reproducibility, inference cost and latency
• Work directly with founders, annotation teams, and client stakeholders. Your judgment on data shapes what the model can do, not just how it gets trained
• Learn how dataset design, labelling quality, and evaluation connect to business outcomes, because on every project the client is paying for outcomes

Key Responsibilities

• Build and maintain data pipelines using Python and SQL.
• Transform raw text into categorical variables and structured features.
• Work closely with annotators to understand labelling guidelines, edge cases and data-quality issues.
• Convert annotated data into formats suitable for model training, inference and evaluation.
• Implement model workflows using predefined architectures and training methods.
• Support recommender systems, including collaborative-filtering approaches.
• Build and optimize inference pipelines for vision and vision-language models such as Grounding DINO and VLMs.

Required Skills

• 4+ years of overall engineering experience with strong software fundamentals. ML engineering on top of weak engineering doesn't work.
• Strong hands-on experience with Python, SQL and data engineering.
• Experience processing unstructured text and engineering categorical features.
• Experience working with annotated datasets and data-labelling teams.
• Practical knowledge of recommender systems and collaborative filtering.
• Understanding of computer-vision and vision-language models from an inference perspective.
• Familiarity with data validation, model evaluation and reproducible ML workflows.
• Strong analytical, problem-solving and communication skills.

Strong pluses

• Hands-on experience with PyTorch: training loops, fine-tuning, and debugging model behaviour
• Working with embeddings and similarity search at scale (pgvector, FAISS, or a dedicated vector database)
• Experience with multimodal datasets: paired image-text data, annotation tooling, and quality control at scale
• Optimizing model inference: batching, quantization, GPU utilization, and latency budgets
• Built internal tooling for data and ML work: annotation dashboards, dataset versioning, evaluation harnesses

AI-native expectations

• Uses Claude Code or Cursor as the primary development environment, shipping production code through AI-assisted workflows daily
• Maintains CLAUDE.md and project context files so AI tools know your conventions, architecture, and constraints
• Plan-first for multi-file changes: AI reads the codebase, you approve the approach, then execute
• Uses AI to accelerate the data work itself: drafting transformation logic, generating test fixtures and edge cases, stress-testing pipelines, triaging model errors
• Tests after every AI-generated change. AI writes fast; you keep it honest

How to apply

Send your resume and a short note to [email protected]. Skip the template. Tell us about something you built — what it did, what was hard about it, and what you'd do differently. That's what we actually want to read.