AI Engineer SDE-III

jhana is an early stage, seed-funded homegrown artificial legal intelligence lab, established at Harvard in February 2022 and made in India. jhana builds datasets, agents, and interfaces that help produce legal work faster and with higher fidelity. jhana’s suite of product offerings use AI to streamline workflows across the legal field including litigation, corporate law and in-house counsel, and the judiciary. jhana was recently declared the most outstanding new legal tech in Asia and Oceania by ALITA, the Asia-Pacific Legal Innovation & Technology Association.

This role will be part of a senior AI development team, tasked with researching, engineering, and launching cutting-edge AI products. This team will be responsible for: working with the product team to design technologies that serve user needs; innovating and creating novel solutions in the legal-tech space; and iterating on these solutions as the userbase and product evolve. All members of this team will be deeply experienced, product- and user-oriented, and up to date with the latest research and technologies in the field.

Our products work with our proprietary corpora of ~13M structured documents as well as with internal documents of clients. Efficient, comprehensive billion-scale search is fundamental to our product. This broader technical problem further interacts with the complexity of the law. For instance, cites-cited-by graphs and graphs of precedent undergoing appeal or overruling inform the relevance of a result. And precedents can be good citation matches for complex reasons, wherein cases from unrelated areas or questions of the law can constitute important precedent. We attempt to excel at these challenges by using simple models for preprocessing, by sequencing traditional and cutting-edge techniques, and by carefully reranking and postprocessing results. This helps deliver generative AI that is anchored to information retrieval and cites and quotes all its claims.

This is an advanced role distributed across research, engineering, and product. Relatedly, it is a combination of invention and optimization. These are the problem statements that this role will likely continue or begin our work on—

Measurement

• Constructing measures of search reliability/comprehensiveness and success/effectiveness; comparing ordered sets of search results; relevant notions of similarity and distance to evaluate IR

• Implementing and automating benchmarks using labeled data and expert human feedback, interfacing with our legal research fellows

• Measures of the popularity of a result for a class of queries, to enable reranking by observing users

Optimization/Scaling/Elasticity

• Indexing algorithms and data structures, and infrastructure for billion-scale, multi-engine search

• Pre- and post-processing hacks, eg. query design

• Segmentation, sharding, concurrency and parallelism, and other clever distribution methods—optimizing latency and time and memory complexity

• Low-dimensional/cost-optimized retrieval methods for enterprise settings

Invention

• Identifying, finetuning, and aligning new vector embedding methods

• Reranking and boosting from data augmentation and live user interaction emit

• Caching mechanisms for high-variance natural language queries

We are a public benefit corporation headquartered in Bangalore. We operate in rapidly changing legal systems with awareness of the stakes at hand. Our intention is to influence beneficence and alignment into the technological systems that are augmenting and replacing human institutions. Our team spans diverse identity and training, from physics and mathematics to law and public policy. We are small, fast-moving, horizontally flat, and built on collaboration between lawyers, academics, and engineers. We ship fast, and every line of code our team writes has a >0.9 expectation of making it to production.

One or more of these might describe you—

• Deep familiarity with DL methods and up-to-date with the latest on Arxiv

• Proficiency with linear algebra and programming tensor math

• Familiarity with complex systems and optimization

• Strong engineering skills, ideally with Python proficiency

• Excited by large-N data, access to compute, and problems that are emergent with scale

• Crave ownership over problem statements

• Hungry for impact and to write code that is actually shipped

• Prioritize career advancement (eg Chief Scientist title) and growing with a high-velocity company more than immediate compensation

Compensation is negotiable based on levels and mutual excitement.

This role will start ASAP and requires in-person presence for at least 4 days of the week at our Bangalore HQ. Full remote is negotiable for superlative candidates who are not India-based.

Come as you are: we are a diverse team constituted by members of different backgrounds in nationality, religion, caste, gender, and sexual orientation. We sincerely and wholeheartedly welcome diverse individuals.