Artificial Intelligence Researcher

About the Role

Valiance builds and deploys AI products for national-scale use cases — large Indian corporates, PSUs, and government bodies. We run our own stack rather than just calling third-party APIs, because at our scale, cost, latency, and control matter.

We're hiring an AI Research Engineer to push the boundaries of what our GenAI/Agentic AI products can do — not research for its own sake, but research that ships. You'll sit at the intersection of applied ML research and product engineering: reading papers on Monday, prototyping on Tuesday, and if it works, it's in production within weeks.This is a high-agency, low-supervision role. You'll be given problems, not instructions

What You'll Do

Performance & Infra Optimization

• Squeeze more throughput out of our GPU fleet — quantization, batching strategies, KV-cache optimization, speculative decoding
• Reduce time-to-first-token and end-to-end latency across inference pipelines
• Optimize vector search performance at scale (ANN algorithms, index tuning, hybrid search, re-ranking) across multi-million document corpora

Model R&D

• Evaluate and integrate new open-source/open-weight models as they release — embeddings, OCR, document classification, reasoning, and generation models
• Design and run rigorous evaluation frameworks (benchmarks, human eval, LLM-as-judge) to decide what actually moves the needle for our use cases vs. what's just a leaderboard win.
• Fine-tune models for domain-specific tasks (e.g., Odia/vernacular language support, industrial safety vision, document intelligence

0-to-1 Product Innovation

• Identify gaps in our current products by working backward from customer pain points, not forward from "what's the latest paper"
• Prototype fast, kill ideas fast, scale the ones that work
• Own initiatives end-to-end: from a research idea to a production feature, without needing to be told the next step

Knowledge Contribution

• Publish research papers / file patents where our work produces genuinely novel IP (this is encouraged and supported, not a side quest)
• Represent Valiance's technical depth in industry forums, benchmarks, and technical evaluations (e.g., government empanelment RFPs where technical credibility matters)

What We're Looking For

• 3-7 years in applied ML/AI, with hands-on experience in at least two of: LLM inference optimization, retrieval/vector search systems, model fine-tuning, computer vision, or evaluation framework design
• Demonstrated ability to read a paper and turn it into working code within days, not months
• Comfortable in a systems-level view: you understand why a model is slow, not just that it is
• Product instinct — you ask "will this make the product better for the customer" before "is this technically interesting
• Strong Python; familiarity with inference frameworks (vLLM, TensorRT-LLM, SGLang, etc.), and vector DBs (Milvus, Qdrant, pgvector, etc.)
• Prior open-source contributions, published papers, or competition rankings (Kaggle, NLP shared tasks) are a strong plus — but production impact matters more than publication count.
• Comfortable working with Claude Code as core daily practice.

Why Valiance

• Real production scale: our AI runs live at Large enterprises & Government — not pilots.
• Own your infra: we self-host models and control our GPU stack, so your optimizations matter directly to margin and speed, not just a vendor's bill.
• Google Cloud Premier AI Partner, Nasscom AI GameChanger and Aegis Graham Bell Award winners
• Direct access to leadership; ideas get evaluated on merit and can go from Slack message to production feature in weeks.