India
Software engineer with 6+ years across production AI, SaaS, geospatial systems, infrastructure and security. Most recent: founding engineer at Klarety AI. Built the production stack from scratch. Agents pull satellite data, write and run code, return real answers. Yale uses it for research. Live at klarety.ai. Two pieces of work I find most interesting from Klarety. First, evidence quality in agent traces. I built rolling context summarization, semantic relevance filtering, and a schema that keeps raw tool output separate from model inference. Query correlation up around 60%. Hallucinations down hard. Second, credential isolation for sandboxed code execution. User code runs arbitrary Python in containers and needs LLM APIs, but real API keys can never enter the sandbox. I built a network-layer proxy that intercepts outbound HTTPS and injects credentials only for whitelisted domains, so even malicious code inspecting env vars or network traffic never sees a real key. Before Klarety, two years at AiDash on Earth Observation SaaS. PostGIS optimization that helped close a $1M contract. PMTiles tiling server doing millions of requests per minute, 50% latency cut, 90% database load cut. Before that, Soroco. REST APIs, microservices on GCP, a fault detection module that cut downtime 45% on Scout. Stack: Python, TypeScript, Node, PostgreSQL, PostGIS, AWS, Docker, CI/CD. Anthropic API for LLM work. Code at github.com/harsh52.
• LLM Development: Designed and implemented an advanced LLM system for a geospatial assistant, including rolling context summarisation, semantic relevance filtering with Sentence Transformers, and token-aware memory management-improved query correlation by 60% thus reduced hallucinations. • Designed and built a zero-trust credential isolation proxy that prevents API key leakage from sandboxed code execution environments. Credentials never enter user-facing containers - a network-layer MITM proxy intercepts outbound HTTPS and injects secrets only for whitelisted domains, reducing credential exposure surface to zero. • Developed a high-performance raster and vector tiling server with full CRUD operation support, capable of handling millions of requests per minute; optimised for superior efficiency and scalability reducing latency by 50% in high-volume geospatial data visualisation and database load by 90% through PMTiles integration. • custom XYZ tile server: Created a custom XYZ tile server with FastAPI and Rust(Martin) for scalable visualisation of raster/vector map layers. • Built and developed Klarety.AI from the ground up in a pre-seed startup, leading initial product architecture and core software.
• Improved existing SaaS services by leveraging PostGIS and Python, enhancing the system by approximately 80%, and assisting the organization in securing a million-dollar contract. • Coordinate with leadership team and implement new feature request in SaaS. • Dockerized and built CI/CD to deploy production code. • Develop and maintain data driven large scale SaaS services. • Developed microservices for seamless integration of a SAAS product with third-party services to enhance traction. • Mentored and lead onboarding of new hires for SAAS product. • Own and ensure alignment of the team learning programs for all members.
•Microservices Integration: Designed and deployed scalable microservices architecture, enhancing system reliability and enabling integration of diverse technology stacks. •End-to-End Development: Led the full lifecycle of REST API development, from design to deployment, tailored to meet complex business needs. •Fault Detection System: Collaborated with product teams to develop a fault detection module for the Scout product, reducing operational downtime by 45% •Agile Methodology: Championed Agile methodologies, resulting in a 25% improvement in project delivery timelines.
•Dashboard Development: Created interactive dashboards using Shiny and React, enabling data-driven decision-making for key stakeholders. •API Integration: Developed and integrated APIs to seamlessly connect front-end applications with backend services. •Code Optimization: Improved code efficiency by 30% through parallel processing techniques, enhancing overall system performance. •Automation Scripting: Automated repetitive tasks using Python scripts, reducing manual workload by 40%.
• Detect different of artifact in medical image. • Analyze different computer vision algorithm on different data-sets. • Write test script to automate task in python.