Abnik Ahilasamy

About

Interned at Larsen & Toubro where they built and deployed an agentic RAG document question-answering system to reduce time spent searching documents and improve trustworthiness. Implemented ReAct-style multi-step orchestration with LangChain/LlamaIndex plus evidence-bounded generation, grounding/citations, and rigorous evaluation—cutting latency ~40%, hallucinations ~35%, and unsafe outputs ~40% while collaborating closely with non-technical business/ops stakeholders.

Experience

Education

Key Strengths

• Built and deployed an agent-based RAG document QA system in production (Larsen & Toubro internship)
• Reduced latency ~40% via caching, batching, and memory reuse
• Reduced hallucinations ~35% using context engineering, grounding checks, and internal citations
• Improved safety by ~40% using faithfulness/safety evaluations and metrics-driven iteration
• Strong agent/workflow reliability approach: stage decomposition, unit tests, adversarial testing, and SLO-based monitoring/rollbacks
• Effective cross-functional collaboration translating business needs into system constraints (grounding, abstention, latency targets)

Reference Highlights

• Took ownership of core technical components
• Strong end-to-end ownership in academic AI projects
• Instrumental in debugging and improving performance
• Able to integrate research ideas into working implementations
• Rigor in experimental setup and analysis
• Strong technical writing; well-presented work
• Emphasis on clean implementation and reproducibility
• Clear, patient, structured communicator
• Effectively translates complex ideas into practical implications
• Proactive in aligning team on goals and dividing work efficiently
• Adapts communication depth to audience
• Clearly communicates assumptions and tradeoffs