Resume

A timeline, not a claim cloud.

Verified experience stays separate from project records. The route keeps the chronology clear without adding unsupported location, work-mode, or employer-project claims.

Publishing rule

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

Timeline

Reverse-chronological experience

Feb 2025 — Present

TapHealth

AI Engineer

Built production clinical-AI systems for diabetes care, including expanded CDSS coverage, lab-aware context pipelines, Tara coaching workflows, and pre-release evaluation and observability loops.

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

Expanded the CDSS from Type 1 and Type 2 diabetes plus a few adjacent comorbidities to 120 modeled conditions: 87 chronic and 33 acute.
Built a lab-context layer covering 45 normalized lab-test parameters across 362 graph nodes and naming variants to support care-plan updates and follow-ups informed by lab results.
Built Tara as a contextual AI health coach across WhatsApp and in-app surfaces; the sourced role record attributes rollout-period growth of daily active users +164%, sessions +187%, and messages +117%.
Built an in-house pre-release evaluation harness with fixtures, personas, rubrics, LLM-as-judge checks, and live validation; Langfuse was used for post-release observability and pre-release prompt experiments.

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

Aug 2024 — Jan 2025

TapHealth

AI Engineer Intern

Worked on the earlier TapHealth health-coach and logging stack, including MCP-based assistant actions, health logging workflows, retrieval, and cache efficiency improvements.

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

Evidence-backed highlights

Built MCP tools and utility MCP servers for assistant actions, structured response generation, backend workflow execution, and health logging workflows.
Supported glucose, food, exercise, weight, and HbA1c logging plus care-plan-grounded retrieval and resource lookup.
Improved macro-fact cache hit rate from 58% to 93%, framed as a direct cache metric rather than a broad model-call reduction claim.

Python · FastAPI · MCP · Retrieval · Caching

Jan 2024 — Aug 2024

Vlippr

Data Science Intern

Worked on multilingual NLP and speech tasks, including IndicBART fine-tuning and in-house evaluation, with diarization claims kept qualitative.

Evidence-backed highlights

Fine-tuned AI4Bharat IndicBART for multilingual paraphrasing.
Reached BLEU 21 on an in-house evaluation.
Worked on diarization, described without unsupported scoring claims.

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