Member of Technical Staff - Post-Training Engineer
About the role
Role focused on post-training of multi-modal models for Terra, building and adapting open-weight models across text, vision, speech, and geospatial modalities, with end-to-end responsibility from data curation to training, evaluation, and deployment in a production setting.
Responsibilities
- Multimodal post-training. SFT and preference optimization on open-weight models across text, vision, and speech — document understanding today; voice agents, video, and geospatial reasoning on the near-term roadmap.
- Agentic behavior tuning. Training models to be good agents—tool use, multi-step reasoning, code-switching across Indian languages, escalation vs. answering, grounding every claim in evidence.
- Provenance-grounded outputs. Structured generation with evidence, schema design, constrained decoding, training strategies to cite rather than hallucinate.
- Speech and voice adaptation. Adapting ASR/TTS and speech-LLM models for Indian languages, accents, and real-world acoustics.
- GIS and visual grounding. Teaching models to reason over survey maps, plot boundaries, and satellite imagery, and reconcile with documents.
- Training data as a product. Annotation taxonomy design, mining production transcripts and failures into training data, building data flywheels.
- Evals that gate releases. Evaluation suites per modality and per agent behavior with regional quality floors.
Qualifications
- 3–7 years in ML engineering with hands-on post-training experience: you've run SFT or preference-optimization jobs on open-weight models and shipped the result.
- Depth in at least one modality beyond text — vision-language, speech, video, or geospatial — with ability to expand to others.
- Fluency with PyTorch, Hugging Face ecosystem (transformers, PEFT, TRL or similar), experiment tracking, GPU training workflows.
- Strong evaluation instincts and ability to design evals before training; explain regressions from error analysis.
- Data-centric mindset with tooling to build curation pipelines.
- Solid Python and engineering discipline to ensure reproducible training.
Skills
About Landeed
Landeed is India's fastest property title search engine
Landeed is building India's (then for rest of the world) fastest and most comprehensive property title search engine. More than 84% of Indian household wealth is stored in real estate, property and land. However, with 2/3rds of all court cases in India dealing with land-related disputes, it is essential that people have accessible public records to both protect themselves against litigation and provide appropriate documentation for facilitating transactions.
More open roles at Landeed
- People & Operations Associate (HR & Admin)1mo agoHybrid (Hyderabad, Telangana, IN) · People & Operations / HR & Admin
- Priority Case Manager1mo agoHybrid (Hyderabad) · Operations
- Property Advisor - II1mo agoHybrid (Hyderabad) · Sales
- Senior GIS Engineer1mo agoHybrid (Hyderabad) · Engineering
- VIP Client Advisor1mo agoHybrid (Hyderabad)