Languages & Frameworks
AI / ML / Data
Cloud & Infra
Databases & Messaging

I'm an SDE based in Hyderabad, India — M.Sc. Data Science from the University of Greenwich (Merit) and Azure DP-203 certified. I build production AI systems: RAG pipelines, event-driven microservices, and cloud data platforms. I recently published a quantitative paper on SSRN showing that tight EU labour markets are positively correlated with gender pay gaps (r ≈ +0.41) — the opposite of what standard economic theory predicts.

Currently open to SDE roles in Germany, Netherlands, and Ireland. EU Blue Card eligible.

📄 Published Research

Why Tight Labour Markets Do Not Close Gender Pay Gaps: Evidence from a 20-Country Eurostat Panel SSRN · May 2026 · ORCID: 0009-0005-4884-1292

20-country Eurostat panel (2019–2024), 11 NACE sectors. Finds r ≈ +0.41 between labour market tightness and gender pay gaps — contradicting competitive equalisation theory. Introduces the Combined Risk Quadrant (HPI × ERS), the first integrated tightness-equity typology in the academic literature. Implemented in WorkforceGuard, an open-source analytics system with SHA-256 hash-chained governance log.

🔨 What I've built

Masova Full-stack restaurant intelligence platform — 6 Spring Boot 3 / Java 21 microservices on GCP Cloud Run behind a Spring Cloud Gateway (JWT HS512, per-route rate limits). 399 API endpoints across order management, payments, logistics, and BI. Event-driven order lifecycle via RabbitMQ with WebSocket delivery to all clients. Dual-write persistence: PostgreSQL (Flyway V1–V8) + MongoDB. EU VAT engine covering 12 countries with fiscal signers (German TSE, French NF525, Italian SDI, UK MTD). 8 Google ADK 1.25 / Gemini agents handling demand forecasting, churn prevention, inventory reorder, and review drafting. 3 production frontends: React 19 web, React Native 0.83 staff app, React Native 0.81 customer app. GDPR Article 17 compliant with data retention policies (2-year customer, 7-year PCI).

WorkforceGuard AI EU workforce intelligence and pay transparency platform underpinning the published SSRN paper. 28-model dbt pipeline over a DuckDB warehouse ingesting 16 Eurostat datasets (LFS, JVS, SES) across EU27. Computes four composite indices — Hiring Pressure Index, Labour Resilience, Equity Risk Score, Transition Readiness — all formula-versioned and audit-traceable. 7 ML models on 32,769 samples; Random Forest achieved 94.7% accuracy and AUC 0.855 on a 912K-record test set. FastAPI backend + React 19 dashboard with evidence packs and governance log for EU Directive 2023/970 compliance audit.

Aequitas UK bus transport policy intelligence platform built as an M.Sc. dissertation (University of Greenwich) and extended into production. A 7-stage validated pipeline processes 274,719 active NaPTAN bus stops, 13,099 BODS GTFS routes, and 1.75M trips across 33,755 English LSOAs (56.5M population) — 103 quality checks, 0 failures. Key findings: Gini coefficient 0.5741 (vs UK income Gini 0.36), Palma ratio 5.702, 4,245 zero-stop LSOAs, 5,189 evening-isolated communities, 612 triple-deprived LSOAs. ML stack: Random Forest (R² 0.472, top SHAP feature: nocar_pct), HDBSCAN clustering, Isolation Forest, 2SFCA accessibility scoring. Production platform: FastAPI + DuckDB warehouse + FAISS RAG chatbot (all-MiniLM-L6-v2 + Gemini 2.5 Flash) + React/TypeScript frontend. 51 analytical sections across 8 policy dimensions, 30 Jinja2 narrative templates.

BillSathi Local-first, privacy-first bill tracking app for Indian households. OCR pipeline: PaddleOCR 2.9.1 primary + EasyOCR fallback (confidence gate 0.60) + 7 vendor-specific parsers (Amazon, Swiggy, Zomato, Blinkit, Zepto, Rapido, generic). Hybrid categorisation engine across 19 spending categories: rule-based → SGDClassifier (HashingVectorizer, partial_fit) → ChromaDB semantic search → Gemini 2.0 Flash Lite fallback, activated by user correction thresholds. 48 API endpoints with a circuit breaker (3 consecutive 429s → 5-min cooldown). Flutter frontend (Riverpod, GoRouter, fl_chart) with spending analytics, price inflation tracking, and Gmail IMAP bill parsing. Deployed on Oracle Cloud Free Tier ARM64 via Cloudflare Zero Trust Tunnel.

💼 Currently — SDE at Innosolv Private Limited (London, remote)

Building two products: an algorithmic trading platform for live NSE equity & derivatives markets (Java 17, Spring Boot 3.2, WebFlux, Zerodha Kite Connect — options analytics engine, Iron Condor strategies, basket orders, HFT module with Bucket4j rate limiting) and Bharat Alpha, an AI-powered Indian equity research terminal with hybrid RAG over 305 annual reports (143K chunks, FAISS + BM25 + cross-encoder, Gemini 2.5 Flash streaming).

📊 GitHub

🎓 M.Sc. Data Science — University of Greenwich, London · Merit · 2022    B.Tech. Electronics & Communication — GITAM, Visakhapatnam · 8.3 CGPA · 2020    Microsoft Azure Data Engineer Associate DP-203 · March 2025

🇪🇺 Germany · Netherlands · Ireland · Austria · EU Blue Card Eligible
linkedin.com/in/souramarti · martisoura@gmail.com · SSRN Paper · ORCID