LinkedIn · Website · Explore my projects ↗
I'm Datta Dhebe, an Information Technology Analyst at NTT DATA with 6+ years of software development experience. I build enterprise backends and practical AI applications, connecting reliable APIs, cloud services, and machine learning with useful product workflows.
My work spans Python, C#/.NET, Java, and Azure. Here, you'll find AI developer tools, incident-memory systems, mobile applications, and reproducible ML experiments.
|
Evidence-backed incident memory Recalls relevant resolved incidents with provenance and supporting evidence for operator review.
Local workflow implemented; cloud integrations await live validation. |
AI tools with workspace context Streaming chat, repository search, Git context, and approval-gated edits and commands with local models.
Includes a VS Code extension and a local development workflow. |
|
Reproducible tabular ML research CatBoost experiments with stratified validation, per-class diagnostics, and auditable model comparisons.
Reference OOF balanced accuracy: 0.94909. Synthetic benchmark; research prototype. |
AI-assisted business follow-up Lead management, reminders, and Gemini-powered reply drafts, with user review before sending.
MVP focused on a practical small-business workflow. |
|
Long-form reading, aloud An Android reader for text, EPUB, PDF and web content, with saved books, chapter controls and background playback.
App screenshots, import limits and setup are documented in the repository. |
Document and developer utilities A web tools hub covering PDF workflows, Base64 conversions, image tools, calculators and email templates.
Feature coverage and static-hosting instructions are documented in the repository. |
Speech model experiments: TTS Humanize · Personal Voice TTS — dataset preparation, Coqui/VITS training, and checkpoint inference.
| Focus | Technologies |
|---|---|
| Backend & APIs | Python · FastAPI · C# · ASP.NET Core · Java · Spring Boot |
| Cloud & data | Azure · Azure Functions · Queues · SQL · MongoDB · SQLite |
| Applied AI & ML | LLM applications · RAG · Ollama · PyTorch · CatBoost · Speech synthesis |
| Web & mobile | Angular · TypeScript · Flutter · REST APIs · Authentication · Role-based access |
- Start with the workflow. Define the problem and make a useful baseline.
- Design for reliability. Make failure cases, permissions, and operator control explicit.
- Measure before improving. Use reproducible experiments and document results and limitations.
Currently exploring grounded AI assistants, agent memory, and local-model tooling.
Let's build something useful.
Open to collaboration on backend systems, practical AI tools, and applied ML.
Connect on LinkedIn ↗