Hi, I'm Harshvardhan Bhosale
Full Stack Developer
Java Developer
Machine Learning Practitioner
Qunatitative Analyst
I build high performance systems that you can actually ship to the users. I leverage microservices architecture with Spring Boot and modern web technologies. Right now, I am building AI bookeeping assistant for construction finance. I am available to work on exciting projects and collaborations.

Experience
My professional journey in development and leadership roles
Software Engineer (Founder)
Mimir
Built and shipped end-to-end real-time VoIP AI agents for customer support, architected multi-tenant versioning for scalability, and integrated advanced conversational AI features.
Key Achievements:
- Built and shipped end-to-end real-time VoIP AI agents for customer support using FastAPI, WebRTC, LangChain, and OpenAI Realtime via Twilio Media Streams, cutting average handle time by 42% and improving first-contact resolution by 28% across 20+ SMB deployments.
- Architected multi-tenant versioning with per-tenant vector stores and rate-isolated pipelines, enabling 5x concurrency scaling while maintaining sub-250 ms round-trip latency for speech-to-action loops.
- Integrated Semantic VAD and intent-aware turn-taking to optimize barge-in and silence detection, reducing interruption errors and boosting CSAT by 18 points through more natural conversational flow.
- Implemented telephonic RAG with tool-use for ticket lookups, order status, and booking flows, automating 65% of repetitive Tier-1 tasks delivering 1.3x ROI within 90 days.
Software Development Intern
Nomura
Developed a dependency upgrade tool, upgraded database instances, and constructed a trade reporting application.
Key Achievements:
- Developed a dependency upgrade tool using TypeScript, orchestrating seamless integration via GitLab pipelines, significantly reducing manual intervention and ensuring consistent updates.
- Upgraded three Oracle database instances, achieving a 40% increase in data processing speed.
- Constructed a robust Spring Boot application utilizing JMS, JDBC, and Oracle SQL for trade reporting achieving a 25% reduction in trade processing time.
- Improved documentation accessibility by scraping Confluence pages to generate vectorized word embeddings, successfully training GPT-based language models that achieved an average BLEU score of 51%.
Software Development Intern
Microsoft
Managed the ingestion and processing of radiology reports and developed solutions for semantic highlighting and text summarization.
Key Achievements:
- Managed the ingestion, caching, and processing of over 1 million radiology reports using Azure Redis Cache, significantly reducing retrieval and processing latency by over 50%.
- Implemented and evaluated solutions for semantic highlighting of critical medical text, substantially enhancing the readability and accuracy of radiology reports.
- Developed NLP-driven text summarization methods, enabling rapid synthesis of medical data from detailed reports, thus accelerating clinical review processes.
Featured Projects
Check out some of my recent work

- Built and shipped end-to-end real-time VoIP AI agents for customer support, cutting average handle time by 42% and improving first-contact resolution by 28% across 20+ SMB deployments.
- Architected multi-tenant versioning with per-tenant vector stores and rate-isolated pipelines, enabling 5x concurrency scaling while maintaining sub-250 ms round-trip latency for speech-to-action loops.

- Architected and implemented a concurrent TCP server in Go, leveraging Goroutines and RWMutexes to efficiently handle simultaneous client connections and maintain high throughput.
- Successfully implemented the REdis Serialization Protocol (RESP) specification, enabling seamless communication with standard Redis clients and supporting core data types (Strings, Arrays, Errors).

- Engineered a full-stack, real-time chat application using the MERN stack and Socket.IO for instant, bidirectional communication.
- Architected a scalable backend with robust APIs and a resilient MongoDB schema to ensure seamless message delivery and persistence.

- Developed a secure RESTful API using Spring Security, featuring robust JWT-based authentication and authorization mechanisms.
- Integrated dynamic, self-updating charts and graphs for intuitive visualization of transaction history and financial data.

- Designed a novel ensemble model combining N-BEATS for time-series analysis with social media sentiment for trend prediction.
- Built a data pipeline to process over 1 million social media posts, using BERT for advanced sentiment analysis.

- Contains solutions to over 200 challenging problems from Codeforces and AtCoder, with a focus on 'hard' difficulty levels.
- Implemented a wide array of advanced data structures from scratch, including Segment Trees, Fenwick Trees, and Tries.
Technical Skills
My expertise across various technologies and tools
Achievements
Recognition and milestones from my technical journey
President of Google Developer Student Club
Convener of University's Technical Team
Top 5 Global Rank at IIT Roorkee's Finance Hackathon
Finalist at Nomura's Kakushin Hackathon
Top 0.8% in JEE Mains 2020 Exam