Anushka04

AI Engineer

Pune, India 2+ yrs exp 86 · Excellent

About

AI engineer with ~3 years building production AI in healthcare: patient-facing voice and chat agents, retrieval over clinical and scientific data, and evaluation for grounding and hallucination. Work spans behavioral telehealth, medical coding (CPT/ICD) and insurance claims, where a wrong answer is costly.

Skills & Expertise (52)

LiveKit Advanced

8.5/10

Years Exp

Python Advanced

8.0/10

Years Exp

LangGraph Advanced

8.0/10

Years Exp

PyTorch Advanced

7.5/10

Years Exp

TensorFlow Advanced

7.5/10

Years Exp

Machine Learning Advanced

7.0/10

Years Exp

Docker Advanced

7.0/10

Years Exp

Deep Learning Advanced

7.0/10

Years Exp

SQL Advanced

7.0/10

Years Exp

Pinecone Unit Testing LINUX Git AsyncIO Django GitHub Gitlab Postgresql MongoDb Supabase Codex ChromaDB Qdrant AWS S3 Redis AWS Azure Medical coding CPT ICD Evaluation ElevenLabs VAPI WebRTC WebSocket Pandas NumPy YOLOv8 OpenCV Hugging Face RAG FastAPI RAGAS Langsmith Langfuse LangChain Autogen MCP n8n Claude Code Mistral DeepSeek

Work Experience

AI Engineer

Storyline Health AI

Mar 2026 - Present

Built a patient-facing care assistant for a behavioral telehealth platform, owning RAG retrieval, agent orchestration (LangGraph), and tool integration over MCP. Built real-time voice (LiveKit, Vapi) alongside chat, grounding answers in each patient’s own records with source attribution. Evaluated answers with RAGAS for faithfulness and retrieval quality, flagging responses not supported by the patient’s records before they reached users. Benchmarked rerankers on NDCG and MRR and selected Cohere Rerank, improving precision@k and lifting answer relevance and accuracy from 78% to 96%. Checked BAA coverage to run Cohere Rerank on models already deployed in our AWS environment, keeping PHI out of third-party APIs. Served a reranker locally with ONNX Runtime for a separate RAG project. Worked closely with the clinicians on the team, who tested the app regularly, iterating on agent behavior and answers based on their feedback.

AI Engineer

Thinking Stack

Feb 2025 - Mar 2026

Deployed a self-hosted reranking stack for a multi-agent medical coding system (CPT/ICD): Infinity inference server on Docker over remote GPU, comparing mxbai-rerank-v2 and bge-reranker-v2-m3, with FastAPI middleware that routes clinical queries and logs ground-truth data for evaluation. Owned reliability and governance for the multi-agent system: LangSmith tracing across every agent, versioned prompts and agent configs, answers checked against an eval set, and custom evals logging user thumbs up/down, human-in-the-loop activations, failure rates and errors, cutting HITL activations from over 30% to under 10% of queries.

AI Engineer

AIQWIP

Sep 2023 - Dec 2024

Built a real-time voice agent on LiveKit, the OpenAI realtime API, and Django, with multi-accent speech recognition around 92%. Debugged the live audio streaming pipeline end to end to bring latency under 500ms, the constraint that makes or breaks voice UX, cutting query resolution time by ~35%. Built a multi-agent LLM system (LangGraph, OpenAI) over large PDF documents with context engineering (embeddings, ChromaDB, chunking, agent memory) reaching ~95% accuracy; evaluated with LangSmith, tracking retrieval quality and hallucination rates.