Rishi R

Built and own end-to-end AI infrastructure (LLMs, vector search, agentic search, ingestion pipelines) powering the AI feature suite at Signeasy at >99% reliability; set the AI architecture direction for the company and review AI design decisions across product and engineering. Lead AI Developer for the flagship Intelligent Contract Management (ICM); designed semantic search combining dense + sparse retrieval, structured extraction, and agentic LLM workflows delivering sub-second p95 query latency over enterprise contract corpora. Designed an agentic multi-tool search system; trained a custom BERT tool-classifier (>99% accuracy, 100-200ms inference) as the primary router with LLM as fallback, eliminating 95% of LLM calls from the hot path and cutting inference cost roughly 10x. Designed large-scale ingestion pipelines (Airflow + Kubernetes) and a Langfuse-backed prompt and eval platform processing thousands of contract PDFs daily, cutting prompt and model iteration time by 80%. Built an end-to-end fraud detection ML system (data, training, deployment, auto-retraining loop) using custom trained XGBoost models; revamped legacy LLM pipelines for summarization and QA.

Joined under contract during final year of B.Tech; received full-time Founding AI Engineer offer post-graduation. Built production AI products from the ground up: an Intelligent Meeting Note Taker and Call Intelligence Analyzer, adopted by enterprise Account Executives (AEs) for post-call workflows. Developed a real-time Voice AI pipeline (Speech to Text to LLM to Speech) achieving sub-second (<1s) end-to-end latency, forming the foundation for their Automated Voice Agents product. Designed RAG-based QA systems and autonomous agents shipped directly to production; automated QC pipelines and NLP features enabled key strategic partnerships and sustained user growth.

Developed a Macro Economic Forecasting Diffusion Index using ML over 70+ longitudinal datasets to drive trends and policy insights.