Resume

AI systems, grounded in practice.

Roles in reverse-chronological order, each with evidence-backed highlights. The route stays separate from project records so chronology never implies an employer relationship.

Publishing rule

Only evidence-backed chronology and role details appear here. Location and work mode are omitted. This web timeline is the current resume format.

Timeline

Reverse-chronological experience

Feb 2025 — Present

TapHealth

AI Engineer

Built agentic clinical-AI systems for diabetes care, spanning a contextual health coach, graph-backed decision support, lab-report interpretation, and pre-release evaluation.

Follows the TapHealth AI Engineer Intern entry immediately below and is represented as a sequential role change without claiming a formal promotion.

Evidence-backed highlights

Built Tara, a contextual AI health coach across WhatsApp and in-app surfaces with Vercel AI SDK, MCP tools, and Hindsight; during rollout, daily active users grew 164%, sessions 187%, and messages 117%.
Built a graph-backed clinical decision support system modeling 120 conditions—87 chronic and 33 acute—to ground personalized diabetes care plans and proactive check-ins with contextual retrieval.
Built a lab-report interpretation pipeline covering 45 normalized lab-test parameters across 362 graph nodes and naming variants, linking them to doctor recommendations and lab-aware Tara follow-ups.
Built an in-house pre-release evaluation harness with fixtures, personas, rubrics, LLM-as-judge checks, and live validation; Langfuse covers post-release observability and prompt experiments.

Python · TypeScript · FastAPI · Kafka · BullMQ · Vercel AI SDK · MCP · Hindsight · Langfuse

Aug 2024 — Jan 2025

TapHealth

AI Engineer Intern

Built an earlier TapHealth health coach and logging stack with LangGraph, MCP tools, retrieval and reranking, and a cache for repeated health facts.

Precedes the sourced AI Engineer entry at the same organization from Feb 2025 onward.

Evidence-backed highlights

Built a LangGraph health coach with MCP tools for glucose, food, exercise, weight, and HbA1c logging, plus care-plan and resource lookup with retrieval and reranking.
Built a macro-fact cache to reduce repeated on-demand Gemini generation; raised the hit rate from 58% to 93% and reduced cache-miss Gemini calls by 83%.

Python · FastAPI · LangGraph · MCP · Retrieval/Reranking · Caching

Jan 2024 — Aug 2024

Vlippr

Data Science Intern

Built a multilingual paraphrasing model with AI4Bharat IndicBART fine-tuning and in-house evaluation.

Evidence-backed highlights

Fine-tuned AI4Bharat IndicBART on internal data for multilingual paraphrasing and achieved BLEU 21 on an in-house evaluation.

PyTorch · TensorFlow · Hugging Face Transformers · Multilingual NLP · Evaluation