Data Science Architect (NLP Expert)

About Neo AI is the most significant shift in how work gets done since the advent of the Internet — yet most organizations fail to capture its value. Neo is changing that. Founded by Bhavin Turakhia — Co-Founder of Zeta, Radix, and Titan — Neo is building an integrated, AI-first suite of products that capture work, centralize context, and make AI a first-class participant in every workflow: • Tasket — A radically reimagined, AI-native work management platform that centralizes context and makes AI delegation seamless for all work. • Friday — AI Assistant, co-worker and agent platform, pre-integrated into Tasket, Studio, Drive and 1000+ SaaS platforms. • Studio — AI-native suite for co-creating docs, spreadsheets and diagrams collaboratively with humans and agents. • Drive — AI-native repository, where humans and agents can co-work on files. • Scribe — AI assistant for meetings that goes beyond note-taking and actually does your work while on the call.

As Architect — Data Science you will own the search and retrieval systems that find, rank, and assemble context across the Neo suite, and mentor engineers across teams. Retrieval is core infrastructure at Neo. It decides what Friday's agents see before they act, what Scribe recalls from months of meetings, and what surfaces when anyone searches across Tasket, Studio, and Drive. It has to be accurate, permissions-aware, and fast. We are proponents of leveraging best-in-breed AI tools at every step of the engineering lifecycle — design, development, review, and testing. This requires a different engineering mindset: a lean team of senior, hands-on, AI-adept engineers, resulting in ultra-rapid iteration cycles and fast output. You will own, oversee, and set the bar for this approach across the retrieval stack, from design through to production. This is a role for someone with genuine research depth in information retrieval and NLP who has carried that work into production, and who is at home reasoning from index internals up to product outcomes.

Retrieval architecture & direction • Own the architecture of search and retrieval across the suite — ingestion and indexing pipelines, hybrid retrieval (lexical + dense), reranking, and query understanding. • Design the retrieval APIs that Friday's agents and the product surfaces call, permissions-aware by construction. • Make the hard-to-reverse calls on index design, freshness, latency, and cost. Search quality & evaluation • Define what "relevant" means for each surface, and build the golden sets and metrics (recall@k, NDCG, MRR) that measure it. • Gate every relevance change behind offline evals and online experiments so quality never regresses silently. Execution & delivery • Take retrieval improvements from research idea to production: write the spec, run agents in parallel to implement, steer them, and verify the result. • Own verification — decide what correct means for the system and make sure what ships meets it. AI-first SDLC harness • Build the harness the team builds with — specs, reusable skills, subagents, evals, and guardrails that keep agent output reliable. • Set the quality gates agent-authored changes pass before a human reviews them. Mentoring & raising the bar • Mentor engineers, run design reviews, and turn what works into standards and paved roads the team adopts. • Spread rigorous relevance-tuning practice across the team, not just your own work. Product thinking • Work with product to pick the retrieval problems worth solving; own outcomes, not tickets. Skills • AI-native engineering practice — Fluent in an agentic development workflow: spec-driven delivery, running and steering coding agents, and building the harness (specs, skills, evals, guardrails) that makes agent output reliable at the team level. • Information retrieval & ranking — Deep grounding in IR: inverted indexes and BM25, dense retrieval, rerankers, and learning-to-rank — with the discipline to tune ranking rigorously rather than chase novelty. • NLP & representation learning — Embedding models, contrastive training, domain adaptation, and fine-tuning; query and document understanding over messy, real-world work data. • RAG & agentic retrieval — Grounding with citations, chunking and context assembly, and multi-step, tool-driven retrieval for agent workflows. • Evaluation & experimentation — Golden sets, offline metrics (NDCG, MRR, recall@k), calibrated LLM-as-judge pipelines, and online A/B testing; regression gating as a habit, not an afterthought. • Search infrastructure at scale — Internals of engines such as Elasticsearch/OpenSearch, Vespa, or Lucene, and vector indexes (HNSW, IVF, quantization); sharding, caching, and holding p99 latency and cost-per-query targets. • Language & runtime fundamentals — Deep proficiency in Python plus one or more of Java, TypeScript/JavaScript, or Go — including concurrency, memory management, and runtime behaviour under load. • Distributed systems — Fault modelling, consistency, and load distribution; the judgment to keep an always-on retrieval layer reliable as data and query volume grow. • Cloud — AWS / Google Cloud / Azure. Experience &