Full-Stack & AI Engineer — Python · FastAPI · Azure · Applied NLP
Building systems that are auditable — where every output can be traced back to the evidence that produced it.
| Status | Open to graduate SWE / AI / Cloud roles — full-time from September 2026 |
| Degree | BSc (Hons) Computer Science, Ulster University London |
| Location | London, UK |
| Focus | Backend services, applied NLP, cloud deployment, verifiable systems |
| Languages | English, Gujarati, Hindi |
Each project below solves one specific hard problem.
🔍 VerifyPulse — real-time news verification
Scores breaking news by how many independent, credible outlets corroborate it — instead of asserting truth, it shows the evidence.
Problem — a single-source story and a ten-source story look identical in a feed.
Approach — RSS + GDELT ingestion → MiniLM sentence embeddings → cosine clustering → 3-factor confidence score (source count, source credibility, source diversity) → FastAPI + live dashboard.
Stack — Python FastAPI sentence-transformers SQLite
Concepts — semantic clustering, weighted scoring design, scheduled ingestion pipelines, deduplication
🔐 SecureTransfer — zero-knowledge file transfer
Client-side encrypted transfer where the server can detect abuse without ever reading a file.
Problem — end-to-end encryption normally kills your ability to detect malicious usage.
Approach — browser-side encryption before upload, server stores ciphertext only, anomaly detection runs purely on transfer metadata (size, frequency, timing).
Stack — Python Flask WebCrypto SQLite
Concepts — threat modelling, zero-knowledge architecture, metadata-only anomaly detection
🧾 Longhand — auditable US-expat tax assistant
Built with one teammate for the AMD Developer Hackathon (ACT II). Every figure the assistant produces links back to the rule and input that generated it.
Problem — LLM tax advice is unusable if you cannot show why a number is that number.
Approach — deterministic rule layer for computation, model layer only for explanation; full citation trail per line item.
Concepts — grounded generation, deterministic/LLM separation, audit trails
Live — amd-hackathon-sepia.vercel.app
☁️ SaaS Usage Monitoring API — multi-tenant metering backend
REST API with tenant scoping, usage aggregation, and rate accounting.
Stack — Python Flask MongoDB JWT
| Layer | Tools |
|---|---|
| Languages | Python, TypeScript, JavaScript, SQL |
| Backend | FastAPI, Flask, REST design, JWT auth, RBAC, input validation |
| AI / NLP | sentence-transformers, TF-IDF, embedding search, semantic clustering, RAG patterns |
| Cloud | Azure App Service, Azure Functions, Netlify, Vercel, Docker |
| Data | MongoDB, PostgreSQL, SQLite |
| Practice | Git, GitHub Actions, Postman, pytest, API testing |
- Extending VerifyPulse into multilingual claim tracking (Phase 3)
- Contributing to google/adk-python — agent development kit
- Dissertation: privacy-preserving transfer with metadata-driven anomaly detection
Open to graduate roles in the UK (Skilled Worker sponsorship required) and, longer term, East Asia.
Portfolio: vishnupro.netlify.app LinkedIn: linkedin.com/in/vekaria-vishnu