Information Retrieval
Hybrid search, learning-to-rank, semantic + lexical fusion tuned to your users and your latency budget.
Boutique AI advisory · BerlinMichelin thinking.
Michelin thinking.
Bistro delivery.
Problem-first. Value-obsessed. We help ambitious teams ship search, RAG, and information-retrieval systems you can measure and defend — not demos that wobble in production.
Founded by Atita Arora — two decades in search, ex-Qdrant, ex-OpenSource Connections.
Talks, articles & podcasts across
Berlin Buzzwords/Haystack/MLOps Community/AI Engineer/Community Over Code/Vector Podcast/DataTalks.Club/Product-Led Hub/Superlinked/OpenSource Connections
What we build
Six practices.
One obsession: relevance.
Every engagement starts with a diagnostic and an evaluation set. From day one you can answer “is this working?” with a number — and every subsequent change is either an improvement or a regression, on paper, before it ships.
RAG Systems
Grounded generation, chunking strategy, evaluation harnesses, hallucination and abstention control.
Relevance Engineering
Judgment lists, offline metrics (nDCG, MRR), online A/B, continuous evaluation pipelines.
AI Strategy
Model & vendor selection, build-vs-buy, latency/cost/quality trade-offs, honest roadmaps.
Vector Search & Embeddings
Choosing vector stores, embedding models, and indexing strategies that actually scale.
NLP
Domain adaptation, extraction, embeddings, fine-tuning where it truly earns its keep.
Recent talks
From the stage.
Product-Led HubJun 2025
From Reactive to Proactive: Is Agentic AI the Future of Decision-Making?
Watch on YouTubeRecent writing
Field notes.
Technical•Mar 2023•1 min
Building Vector Search in Chorus — a Technical Deep Dive
A deep technical writeup on adding vector search capabilities to Chorus — the open-source search-relevance workbench — including architecture, indexing, and demo workflows.
vector-searchchorusopensource
Technical•Mar 2023•1 min
Revolutionizing E-commerce Search with Vectors
How vector search reshapes e-commerce retrieval — semantic similarity, natural-language queries and the honest limits of pure dense retrieval.
vector-searchecommercewriteup
Technical•May 2022•1 min
Make Quepid Talk to Your .NET Search API
A step-by-step guide to wiring Quepid — the search-relevance judgement tool — to a .NET search API for offline evaluation.
quepidevaluationdotnet
Discuss a project
Have a hard search problem?
We take on a small number of engagements each quarter. Tell us what you’re building and we’ll tell you honestly whether we’re the right fit.