4 projects · C#/.NET · Python
Projects
A foundation-discovery pipeline, a stock-market data platform, enterprise apps and a memory for agent teams.
| Project | Problem and what I did | Outcome |
|---|---|---|
| Grant HunterGPT · Serper · Firecrawl · Prometheus · Grafana | Foundations processed and enriched from multiple sources. ProblemFoundations found by an AI-driven workflow had to be enriched and validated at scale. DidDesigned and implemented the high-throughput pipeline; GPT workflow with Serper and Firecrawl to enrich, cross-check and validate; logging and observability with Prometheus and Grafana. | 100,000+foundations |
| HamroLaganiPython · web scraping · API | Nepalese stock-market data scraped, processed and served to applications through an API. ProblemNepalese stock-market information had to be collected from the web and made usable by applications. DidBuilt the Python scraping and data-processing workflows, analysed the collected data, and built the API that serves it. | Data platform |
| SBSC & OMSC# · ASP.NET Core · SQL Server · cloud · GenAI APIs | Two enterprise applications, from the backend to the database. ProblemEnterprise applications needed backend services, a database layer and AI integrations. DidImplemented backend functionality and API services in C# and ASP.NET Core; designed SQL Server structures and stored procedures; contributed to cloud development and generative-AI API integrations. | Enterprise apps |
| KnewatFalkorDB · MCP · Cypher | Rule checks in two cross-session trials: 0 / 36 broken with Knewat, against 33 / 36 for plain Claude Code (4 runs per arm, no LLM judge). ProblemAgent harnesses remember in text files and coordinate through task lists, so a team of agents has no shared memory. DidBuilt one shared graph for agent teams: memories, tasks, claims, handoffs and decisions are nodes, and agents connect over MCP. Each tenant gets its own graph. Source →Write-up → | 0 / 36rule violations · 2 trialsvs 33 / 36 plain |