Sr Software Engineer - Coding Agents

Story Terrace Inc.; Mumbai; On-site; INR 3,000,000 - 5,000,000; Full time

Job summary

LexsiLabs is seeking a developer tooling specialist to lead the agent tooling surface for software work, including retrieval, editing, builds, tests, static analysis and version control. The role emphasizes verification of changes before human review, long-horizon execution with checkpointing, and robust integration with engineering environments.

You will work with our research team on evaluation design, post-training and interpretability of agent behavior.

Job description

LexsiLabsistheleadingfrontierAIlabfocusedonbuildingaligned,interpretable,andsafesuperintelligentsystems.Whilethatisthevision,themissiontobuildsafetyawareautonomoussystemintheextremenearterm.OurresearchworkspansaroundareaslikeAIalignmentmethodologies,interpretability ledsystemdesign,andfoundationalmodelresearchacrossstructured,tabular,andnewautonomoussystemdesigns.Wepublishedabout25+papersinthepast15monthsacrossleadingconferencesICLR,ICML,WWW,IJCNN,MICCAI,Euripsetc.OurlabsarelocatedinIndia(Mumbai&remote),Paris,&London.

Weoperatewithaflatstructure,highautonomy,andastrongbiastowardengineerswhotakefullownershipofwhattheybuild,fromarchitecturetoproductionbehavior.

TheRole

Ourcurrentsprintonbuildingautonomoussystemforcomplexproblems,acrosssoftwareengineering,datascience,andAIresearchinvolvesbuildingharness,executionsubstrateandevaluationsystem,andeachisdesignedtoruninsideacustomer'senvironmentratherthanours.

Thisroleownsthecodingagent:asystemthattakesanobjective,worksdirectlyonarepository,verifiesitsownchanges,andproducesarecordofwhatitdid.Thecodebasesarelarge,old,thinlytestedandload-bearing,andtheagentrunsinsidethecustomer'snetwork,usuallywithoutinternetegress.

What'llyouworkon:

• Theagentarchitectureanditstoolsurfaceforsoftwarework,includingretrieval,editing,builds,tests,staticanalysisandversioncontrol.
• Verification,sothatachangeisestablishedascorrectbeforeahumanseesit.
• Long-horizonexecutionacrossmulti-filerefactorsandmigrations,withcheckpointing,recoveryandself-correction.
• Theauthorisationandrollbackmodelgoverningwhattheagentmaydoandhowitisundone.
• Integrationswiththeengineeringestatetheagentoperatesin,andtheobservabilitytoolingusedtodebugandauditit.
• Evaluationsuitesbuiltfromrealrepositorytasks,andthebehaviortuningthatfollowsfromthem.

Youwillworkcloselywithourresearchteamonevaluationdesign,post-trainingandinterpretabilityofagentbehavior.

WhatWeAreLookingFor

ThisisadevelopertoolingroleasmuchasanAIrole.Theagentisonlyascapableasthetoolsandintegrationsitisgiven,sodepthheremattersmoretousthananythingelse.

• DeepexperiencewithdevelopertoolingandIDEs.EditorandIDEinternalsorextensions,languageserversandLSP,codeintelligence,refactoringengines.Youhavebuiltthingsengineersuseeveryday.
• Codeintelligenceandprogramanalysis.ASTsandtree-sitter,staticanddataflowanalysis,symbolindexingandcodesearch,callgraphanddependencyresolution,codemodsandautomatedmigrationtooling.
• BuildsystemsandCI/CD.Bazel,Gradle,Maven,Makeandsimilar,incrementalandhermeticbuilds,artifactanddependencymanagement,pipelinedesignacrossJenkins,GitLab,GitHubActionsorequivalents,includingself-hostedandofflinedeployments.
• Observability.Distributedtracing,structuredlogging,OpenTelemetry,andbuildingthereplayandinspectiontoolingthatmakesnon-deterministicsystemsdebuggable.
• Integrationsengineering.GithostingAPIs,ticketingandworkflowsystems,enterpriseauth,andtherealityofworkingagainstself-hostedsoftwareseveralversionsbehindcurrent.
• Agenticsystemsinproduction,notprototypes.Youknowthecurrentframeworklandscape(ReAct-styleagents,LangGraph,LangChain,SemanticKernel)andwhereeachonestopsworking.
• Strongbackendfundamentals.AdvancedPython,APIsandservices,sandboxingandcontainerisation,databasesanddatapipelines,cloudandon-premiseinfrastructure.
• Youreasonaboutperformance,reliability,cost,safetyandinterpretabilityasoneconnectedsetofconstraints.
• Youmakegoodcallsinthegrayarea,andyouarecomfortablewhenproblemsarelooselyspecifiedandownershipisassumedratherthanassigned.
• ExperiencebuildingevaluationdatasetsorframeworksforAIsystemsisaplus.
• Compilerbackground,largemonorepowork,orlegacymodernisationprogramsareaplus.
• Priorexposuretoalignment,interpretabilityorsafetytoolinginproductionsettingsisaplus.

Wemovequicklyandexpectcandidatestodothesame.Wevaluesubstanceoverpolishandexecutionoverrhetoric.