Case studies

Legal-tech RAG

multi-tenant retrieval with citations

Built the document ingestion, embeddings pipeline, and citation-anchored RAG chatbot for an AI legal-tech startup. Multi-tenant across orgs, lawyers, and cases. Migrated the vector store from Pinecone to a self-hosted Elasticsearch cluster on GCP.

RAG, Embeddings, Elasticsearch, Pinecone, Multi-tenant

Ad Platform Rebuild

10× revenue over two years

Moved Cafebazaar's ad placement from a brittle monolith to a microservices platform without downtime. Architecture, pricing, ad-placement A/B testing, ML-model A/B testing, and graceful ML fallbacks all contributed - the headline outcome was 10× ad revenue sustained over two years.

Microservices, Python, Go, Kafka, Redis

Kafka + Redis Event Pipeline

50K events/sec

Built the event pipeline behind the ad platform - Kafka topics, Redis-tracked deduplication, batched DB writes. At-least-once delivery with idempotency keys gives effectively exactly-once at the application layer.

Kafka, Redis, Python, Distributed Systems, Infrastructure

Real-Time Estimation

700ms p99 → 25ms

Cut p99 on the real-time ad estimation service from 700ms to 25ms - same endpoint, more traffic - through multi-tier caching, async refresh, request deduplication, batch inference, and graceful fallback when the model service was degraded.

Machine Learning, Python, Redis, System Design, Performance

Company-wide Feature Flags and A/B Testing at Divar

Designed and built the feature-flag and A/B testing service used across Divar's engineering teams - proposed to the CTO, approved, adopted org-wide, kept as standard.

Python, Go, System Design, Backend, A/B Testing

P2P Payment Fraud Detection at Cafebazaar

Rule-based fraud scoring on the peer-to-peer payment platform - cut manual review workload by ~70% by auto-deciding the obvious cases and prioritising the rest.

Python, Django, Fraud Detection, Payment Systems, Backend