LL
Full time
🤖 AI / ML
Research Engineer — Agent Architectures (Coding & Autonomous Systems) at Lexsi Labs
Listed by DevShelfHub
Mumbai, India & Remote; labs also in Paris and London
WFH
Senior · 5–9 yrs
About the Role
Frontier AI research is accelerating, and Lexsi Labs is building aligned, interpretable, and safe autonomous systems. Lexsi Labs is hiring a Research Engineer focused on Agent Architectures for coding and autonomous systems—researching the system around the model, not just the model itself.
This role works on harness design, memory schemas, and evaluation frameworks for coding agents that run inside customer networks against large, legacy, thinly-tested codebases.
What you'll do
- The coding agent is the primary testbed—working on repositories with no internet egress under real latency and cost budgets.
- Harness: action space and tool surface design, context construction policy, control topology, verification design
- Memory: schemas for execution state, compaction policy, retrieval under closed-world constraints
- Evals: task construction from repository history, scoring long-horizon work, variance and contamination control
- Run controlled experiments on architecture—not prompt tuning
- Build what experiments need and distinguish real gains from setup artifacts
- Work with alignment researchers on post-training, behavioral evaluation, and trace inspection
- Publish papers, benchmarks, and shipped architecture
What we're looking for
- 5+ years in research engineering, systems, or ML infrastructure; 2+ years on agentic/LLM systems
- PhD in ML, PL, systems or equivalent industrial research work
- Public record: publications, open-source project, benchmark, or substantial technical writeup
- Two or more years building agents against real workloads (ReAct, LangGraph, LangChain, Semantic Kernel)
Skills & Technologies
Required
5+ years in research engineering, systems, or ML infrastructure; 2+ years on agentic/LLM systems
PhD in ML, PL, systems or equivalent industrial research work
Public record: publications, open-source project, benchmark, or substantial technical writeup
Two or more years building agents against real workloads (ReAct, LangGraph, LangChain, Semantic Kernel)
Built an eval dataset or harness used by others; SWE-bench-class benchmark experience
ASTs, tree-sitter, static and dataflow analysis, symbol indexing, call graph resolution
Advanced Python; sandboxing and containerisation; distributed parallel trial execution
Distributed tracing, structured logging, OpenTelemetry for debugging non-deterministic systems
Benefits & Perks
- Health Insurance
- Flexible Leave Policy
- Learning Budget
- EPF / NPS
Why This Role is Good for Experienced Professionals
- This is a research engineering role where architecture experimentation—not prompt tuning—drives agent performance improvements.
- Research harness design: action space, tool surface, context construction, and control topology
- Build memory schemas for execution state outside the context window
- Design eval datasets from real repository history with executable verification
- Run controlled experiments: isolate failures, form hypotheses, ablate, and validate gains
- Work directly with alignment and interpretability researchers on post-training and behavioral evaluation
- Publish papers, benchmarks, and released tooling
- Build repository-scale indexes using ASTs, tree-sitter, and static analysis
- Deploy sandboxed, containerized experiments at scale (Docker, gVisor, Firecracker)
- Apply post-training methods (SFT, DPO, RL for agents) and inference-time scaling
- Join a lab with 25+ publications at top ML venues in 15 months
About Lexsi Labs
Lexsi Labs is a frontier AI lab focused on building aligned, interpretable, and safe superintelligent systems. Research spans AI alignment methodologies, interpretability-led system design, and foundational model research. The lab has published 25+ papers in the past 15 months across ICLR, ICML, WWW, IJCNN, MICCAI, and NeurIPS. Lexsi operates with a flat structure, high autonomy, and a strong bias toward engineers who take full ownership from architecture to production behavior.
- Industry
- Technology
- Company Size
- 500+
- Website
- lexsilabs.com
🚀 Boost your chances
Skip the queue
Contact the recruiter directly at Lexsi Labs and follow up personally — most applicants never do.
Browse Recruiter DatabaseFree trial · 40 contacts · ₹0
Job Details
- Type
- Full time
- Level
- Senior
- Experience
- 5–9 years
- Location
- Mumbai, India & Remote; labs also in Paris and London
- Work Mode
- WFH
- Category
- AI / ML
Share this Job
Recruiter Database
Want to stand out? Talk to the recruiter directly.
Most applicants never hear back. Our Recruiter Database gives you direct email access to the people making hiring decisions — so you can follow up personally and actually get noticed.
- Recruiter name, company & direct email address
- Sourced from active job postings across top companies
- Free trial — 40 contacts, no credit card needed