ML Lead / Principal Research Scientist - Domain Foundation Models
Software Engineering, Data Science
Bengaluru, Karnataka, India
(PhD Preferred | Founding Engineer Role)
Location: Bangalore India
Type: Full-time | Founding Leadership Role
Why This Role Matters
H2LooP is building a category-defining AI platform for system engineers — not generic copilots, but engineering-grade foundation models that understand specifications, constraints, and real hardware behavior.
Our core team comes from Google, Toshiba, Cisco, Bosch, Philips, and NXP, bringing deep experience across AI, semiconductors, automotive, telecom, and embedded systems.
As a Founding Engineer, you will shape the architecture, technical direction, and engineering culture of the company from day one. This is a rare opportunity to define how domain foundation models are built — and see them deployed in real engineering environments.
Role Overview
We are looking for a ML Lead / Principal Research Scientist to own the data, evaluation, and learning strategy for domain-aware foundation models powering embedded, automotive, UAV, and IoT development.
This role sits at the intersection of machine learning research, system engineering knowledge, and real-world deployment. You will turn complex engineering artifacts into learning signals that enable models to reason about specs, constraints, safety, and execution behavior — not just generate statistically plausible code.
If you enjoy foundational research with direct industrial impact, this role is for you.
What You’ll Build & Own
1. Data Strategy for Domain Foundation Models
Define the end-to-end data strategy for training domain-aware foundation models.
Design domain vs. general data mixtures to balance specialization and generalization.
Establish curriculum-style learning progressions aligned with real engineering workflows.
Own data versioning, lineage, and reproducibility across model iterations.
2. Data Preparation & Knowledge Structuring
Lead pipelines that transform complex technical artifacts into model-ready representations.
Align engineering documentation with code, behavior, and constraints.
Drive approaches for semantic structuring and retrieval of engineering knowledge.
Guide synthetic data strategies to expand rare but critical engineering patterns.
3. Domain-Specific Intelligence
-
Standardize heterogeneous engineering inputs:
Technical documentation
Embedded and firmware codebases
Open engineering stacks and SDKs
Preserve implicit engineering logic such as timing, safety, and system constraints.
Curate negative and failure-mode datasets reflecting real-world engineering mistakes.
Maintain high-quality datasets from open ecosystems (RTOS, middleware, control stacks).
4. Instruction & Workflow-Aligned Datasets
Design instruction and multi-turn datasets reflecting real engineering problem-solving.
-
Define task families such as:
Specification-to-code translation
Code repair and validation
Configuration, integration, and debugging
Establish quality, correctness, and compliance filters aligned with engineering standards.
Ensure tight integration with developer tools and workflows.
5. Feedback-Driven Learning & Evaluation
Define strategies to incorporate execution feedback and system signals into learning loops.
Work closely with system engineers to integrate human feedback into training datasets.
Build domain-specific evaluation benchmarks beyond generic LLM metrics.
Track metrics that reflect correctness, recall, safety, and engineering reliability.
Who You Are
Required
PhD (or equivalent research depth) in ML, Systems, CS, or a related field.
Strong background in data-centric ML or foundation model development.
Experience working with complex, technical, or domain-heavy datasets.
Ability to operate in high-ambiguity, early-stage environments.
Passion for building systems that move from research → production → real users.
Preferred
Exposure to embedded systems, automotive, robotics, telecom, or safety-critical software.
Experience designing evaluation methodologies for domain-specific models.
Familiarity with engineering constraints, standards, or system-level reasoning.
Why Join H2LooP
Build foundational technology, not incremental tooling.
Define a new category of AI for system engineering.
Work with a team that has shipped systems at Google, Bosch, Cisco, Philips, NXP, and Toshiba.
See your work deployed on real hardware, real codebases, and real engineering teams.
High ownership, deep technical autonomy, and direct influence on company direction.
