About Faro
Faro is an AI memory layer that allows agentic systems to recall what it learned in the past. Faro currently outperforms industry leader Mem0 on the LoCoMo benchmark (94.1% vs 91.1%). Faro is the memory infrastructure that makes every AI agent more engaging and accurate to work with.
The Role
As an ML/AI researcher at Syntra, you will own the research geared at improving the Faro engine across areas such as: retrieval strategies, memory extraction quality, temporal reasoning, and the evaluation methodology.
You will design benchmarks, build the datasets and LLM judges that score them, run controlled experiments, and turn your research into production ready features. You will play a key role in keeping Faro ahead of its competitors and helping shape the overall product direction.
What you will Work On
Benchmark datasets, ground truth, and evaluation harnesses
LLM-judge scoring pipelines with confound controls
Retrieval, chunking, and memory-extraction experiments
Accuracy and latency improvements to the core recall engine
Written recommendations backed by powered, reproducible measurements
Research code that merges to main in small, production-safe PRs
Must-Haves
3+ years of applied ML/NLP research or equivalent industry R&D
Strong Python capabilities
LLM pipelines: prompt design, structured extraction, LLM-as-judge evaluation
Embeddings + vector search (Qdrant, pgvector, Pinecone, or similar)
Benchmark design: dataset construction, ground truth, statistically powered comparisons
Engineering discipline: pytest, CI, code review, small reviewable PRs
Self-starter or experienced with "startup pace"
Nice-to-Haves
Memory-systems or RAG research (contextual retrieval, temporal reasoning, consolidation)
Multi-provider LLM experience (OpenAI, AWS Bedrock, Cohere embeddings)
Agent framework experience (LangChain, LlamaIndex, CrewAI, AutoGen)
Publications or open-source evaluation and benchmark contributions
Early-stage startup experience (Series A or earlier)
