engineering.
Currently working as a Founding Engineer at DocuraHealth (YC W26). Previously, I worked as a Machine Learning Engineer at Pibit.ai (YC W21), and a Founding Engineer at Aarogya ID.
+
-
DocuraHealth (YC W26)
Founding Engineer
- 0 -> 1
- #1 Founding Engineer
- Built and improved a Planning Agent system for report generation workflows, orchestrating extraction planning, sample coordination, report preparation, and iterative refinement/replanning for failed or low-confidence outputs, reducing manual operational effort by ~50%.
- Developed Verification Agent layers across multiple AI pipelines to validate extracted data and generated reports, reducing report inconsistencies and manual review/correction workload by ~40% for operations teams.
- Worked on Context Handling systems to improve reasoning consistency and structured output reliability, leading to noticeably higher report quality and reduced regeneration cycles.
- Optimized report generation and Review of Records (ROR) workflows using asynchronous processing and parallel execution strategies, decreasing processing latency by ~60% and significantly improving throughput for large-scale healthcare record handling.
- Enhanced OCR and document extraction pipelines by improving prompt behavior, extraction orchestration, and validation logic, improving extraction accuracy and reducing downstream correction effort.
- Integrated and onboarded new healthcare client workflows, ensuring generated outputs matched client-specific templates, formatting standards, and reporting requirements while reducing template adjustment/review effort by ~50%.
- Worked on Phone Agent workflows capable of processing call/audio inputs and converting extracted information into structured healthcare forms, reducing manual form-filling effort for operational workflows.
- Reverse engineered a proprietary .ds2 (Olympus CELP-based) audio format and built a custom decoding pipeline (DS2 → WAV → OGG Opus) to enable accurate transcription using Deepgram/Whisper, overcoming lack of native FFmpeg support.
- Led end-to-end development of AI infrastructure, workflow automation systems, frontend upload/review flows, and production healthcare AI pipelines as the primary engineer.
- Tech Stack: Claude Opus 4.7, GPT-5.3, Python, FastAPI, Docker, LlamaIndex, Deepgram, Whisper, Next.js, AWS, Async Processing, OCR Pipelines, Multi-Agent Systems, Prompt Engineering, AI Verification Systems, Audio-to-Structured-Data Pipelines, Report Generation Workflows.
+
-
Pibit.ai (YC W21) - Series A
Machine Learning Engineer
- Developed a Python-based comparison system using Pandas and Openpyxl to validate model outputs against ground truth data. The system automatically highlighted discrepancies and critical issues, enabling QA and Operations teams to reduce manual review effort by ~70% and generate actionable reports instantly.
- Improved the performance of the Exmod LLM system by iterating on prompts, evaluating model behavior, and analyzing responses through Langfuse, resulting in a ~25% improvement in document extraction accuracy and reliability.
- Identified and resolved data contamination and document quality issues across multiple workflows, improving data integrity and downstream extraction performance.
- Enhanced supplemental document processing pipelines through prompt optimization and backend workflow improvements, reducing manual correction and QA effort by ~40%.
- Collaborated with cross-functional engineering teams to develop the Piparse package, standardizing document ingestion, preprocessing, and extraction workflows across multiple client pipelines.
- Contributed to the design and development of an internal prompt management system for centralized prompt versioning, evaluation, and experimentation across large-scale LLM workflows.
- Tech Stack: GPT 4o, Python, Pandas, Openpyxl, Prompt Engineering, Langfuse, AWS S3, AWS CloudWatch.
+
-
Aarogya ID
Founding Engineer
- 0 -> 1
- #1 AI Guy
- Designed and evaluated a medical question-answering system by benchmarking multiple LLMs including Claude Sonnet 3.5, OpenBioLLM, and Meditron, identifying the most accurate and contextually relevant models for production use.
- Engineered and optimized document digitization workflows using AWS Textract and Claude 3.5, reducing extraction errors by ~40% across medical documents.
- Integrated Gemini Flash 2.0 into backend digitization pipelines, improving parsing performance and increasing throughput by ~60% on large PDF files.
- Led the backend integration of Gemini Flash 2.0, handling complex API responses and workflow orchestration using AWS Lambda, Bedrock, and CloudWatch.
- Migrated backend storage from AWS RDS to DynamoDB, reducing infrastructure costs while maintaining system performance and scalability.
- Designed and deployed REST APIs through AWS API Gateway, enabling secure and scalable access to document digitization services.
- Authored technical documentation covering system architecture, LLM evaluations, API design, workflow automation, and operational handoff processes.
- Worked closely with founders and product stakeholders to evaluate model performance, improve workflow reliability, and support the evolution of AI-powered healthcare infrastructure.
- Tech Stack: Claude Sonnet 3.5, OpenBioLLM, Meditron, Gemini Flash 2.0, AWS Lambda, AWS Bedrock, AWS CloudWatch, AWS API Gateway, AWS Textract, DynamoDB, AWS RDS, Python, REST APIs.
+
-
Multiple Startups
Engineering
2022 - 2024
India
- While in college: worked with 5 startups (including AarogyaID), built & shipped software, taught Python programming to students and educators at USF , and helped start an Open Source organization at my college.
- From full-stack applications and Machine Learning to RAG/LLM systems. I took things from idea → code → production.
- Tech Stack: LangChain, GPT-4.1, Scikit-learn, NLTK, BeautifulSoup4, Selenium, Python, Flask, Django, React, PHP, JavaScript, SQL, XAMPP, Git, GitHub, AWS EC2.
other stuff.
- Teaching: Taught Python Programming at USF for a few months. I also used to teach my friends back in my college days right before engineering exams and vivas. I love teaching—it helps me gain more knowledge. If you think I can teach you anything, don't overthink it and email me.
- Technical Sessions: I like conducting technical sessions (remote or online) and have covered topics like Building AI Agents, Understanding MLOps Basics, & MLFlow for experiment tracking. If you'd like me to conduct a session for your college or club, reach out to me.
- Opensource Organization: Helped start the Open Source organization at my college (although the college later halted it).
- Football: From age 8 to 13, I used to play football like Lionel Messi—my dribbling was top-tier. Only the people from my old neighborhood will vouch for me on this, but I hope they remember!
- Academic Debates: During my Polytechnic project, a senior professor argued that Python was strictly for the frontend. When I explained to him that it is not, his exact words were: "Don't teach me about frontend and backend." I still remember his face.
toolkit.
Domain Proficiency: Machine Learning, AI Agents, Multi-Agent Systems, RAGs, Evaluation, MLOps/DevOps, Backend Development, Full Stack Software Development
Frameworks & Libraries: HuggingFace, Langchain, LangGraph, Pydantic, LlamaIndex, Crew AI, Phidata, TensorFlow, PyTorch, Keras, Scikit-learn, Numpy, Pandas, NLTK, Spacy, Flask, FastAPI, Giskard, DeepEvals
Tools & Platforms: Git, GitHub, Claude Code, Cursor, Opencode, Codex, Docker, LangFuse, Langsmith, DVC, MLflow, Postman, DataStax, Unsloth
Databases: MySQL, ChromaDB, Pinecone, FAISS, DynamoDB, MongoDB
Cloud & Infrastructure: CI/CD, Docker, Kubernetes, HELM, ArgoCD, Github Actions, AWS EKS, AWS Bedrock, AWS Lambda, AWS S3 Bucket, AWS RDS, AWS DynamoDB, AWS API Gateway, AWS CloudWatch, AWS EC2
Additional Skills: Next.js, PHP, React & React Native, Teaching