Senior ML/AI Engineer

LOCATION

India (Remote)

TYPE

Full-Time

EXPERIENCE

4 – 7 Years

FUNCTION

ML / AI Engineering

Recruit Signals is an AI-powered hiring intelligence platform that helps recruiting sales professionals identify companies most likely to hire in the near term. We analyse a proprietary set of signals drawn from multiple data sources to produce a predictive score for thousands of companies — surfacing the right targets at the right time.

We are a lean, ambitious team building at the intersection of machine learning, LLMs, and B2B sales intelligence. Our data infrastructure is in place (LinkedIn data via vendor, in-house data engineer) — we are now hiring the core ML/AI brain of the product.

You will be the primary ML/AI engineer for Recruit Signals, owning the intelligence layer end-to-end. This means taking clean, structured data from our vendor pipeline and turning it into accurate, explainable, and actionable hiring predictions.

You will not be building data pipelines from scratch — our in-house data engineer owns that layer. You will work closely with them to specify the features and data structures you need, and focus your energy on what matters: building models that are genuinely predictive, and AI systems that extract signal from unstructured content.

• Design and train sequential models (GRU/LSTM) on time-series company data to predict hiring spikes 1–2 months in advance

• Engineer features from structured data across multiple signal categories covering company activity, talent movement, and market signals

• Build and maintain multiple model pipelines suited to different company profiles, each scoring independently based on relevant signals

• Handle class imbalance, sliding window construction, temporal train/test splits, and model evaluation using ranking-focused metrics appropriate for the use case

• Own model versioning, experiment tracking (MLflow or equivalent), and regular retraining cycles

• Experience with LinkedIn data specifically — either via enrichment APIs (e.g., Proxycurl) or third-party data vendors

• RAG architecture design and implementation

• MCP (Model Context Protocol) server development — increasingly relevant as we build a conversational interface layer

• Experience at a B2B SaaS, HR tech, or sales intelligence company

• Familiarity with TimescaleDB, DuckDB, or similar analytical/time-series databases

We are deliberately early-stage and lean. You do not need experience with:

• Distributed computing (Spark, Hadoop) — not needed at our current scale

• GPU cluster management or MLOps at scale — cloud instances handle this

• Frontend development

• DevOps or infrastructure ownership

If you have these skills, great — but they are not the reason we are hiring you.

• 4 to 7 years of hands-on ML/AI engineering experience — you have built and owned systems, not just contributed to them

• Comfortable with ambiguity and capable of making architectural decisions independently without a tech lead above you

• You think in terms of outcomes (does this model actually improve precision?) not just outputs (the model trained successfully)

• Able to communicate clearly with non-technical stakeholders — explaining why a company scored 87 in plain English matters as much as the model accuracy

• Based in India, available to work in a remote-first environment with regular async collaboration across time zones

Salary

Competitive, benchmarked to senior IC roles in the Indian market

Work Style

Fully remote, async-first

Ownership

High autonomy — you own the intelligence layer of the product

Growth

Early hire at a fast-moving AI-native product; direct line to founding team

• Build LLM-powered signal extraction pipelines — identifying hiring intent signals in LinkedIn company posts, job descriptions, and press releases

• Develop prospect identification logic: given a hiring signal at a company, surface the right person to contact (role, seniority, department)

• Apply NLP techniques — NER for location/entity extraction, text classification, semantic similarity — to extract structured signals from unstructured content

• Integrate embedding models and retrieval (RAG) as the product evolves toward a conversational copilot interface

• Work closely with the data engineer to specify feature requirements — you define what you need, they build the pipeline reliably

• Translate model outputs into explainable, human-readable signals that make it immediately clear to users why a company has scored highly

• Contribute to product decisions on model design, output calibration, and handling of edge cases in the data

• Hands-on experience training sequential models — GRU or LSTM — in PyTorch or TensorFlow

• Strong time-series feature engineering: rolling averages, lag features, growth rate computation, sliding window datasets

• Experience with class imbalance techniques (class weights, SMOTE) and ranking-focused evaluation metrics (Precision@K, AUC-ROC)

• Familiarity with MLflow or equivalent for experiment tracking and model versioning

• Practical experience with LLM APIs (OpenAI, Anthropic Claude, or similar) for production use cases — not just experimentation

• NLP fundamentals: Named Entity Recognition, keyword extraction, text classification

• Embedding models and semantic search / retrieval — building systems that find relevant entities or passages at scale

• Prompt engineering for structured output extraction from noisy, real-world text

• Strong Python — clean, well-structured, production-ready code

• SQL proficiency: writing efficient analytical queries, working with materialized features, understanding incremental data patterns

• Ability to work with structured data from third-party vendors and specify feature requirements clearly for a data engineering partner

• Intro call — 30 mins with the founding team to discuss the role and your background

• Technical screen — take-home or live: feature engineering on a sample dataset + model evaluation discussion

• AI engineering challenge — LLM-based signal extraction from a realistic sample of LinkedIn/job posting text

• Final round — architecture discussion: how would you approach building a predictive scoring system for this type of problem end-to-end?