Akshata Vijay Kulkarni

About

Full-stack engineer focused on productionizing LLM applications, including an Android privacy-policy risk summarization app (Kotlin/React Native + FastAPI + Ollama) that cut response times from ~10s to ~5–6s via batching, caching, async, and event-driven architecture. Currently at PRGX building an LLM-based legal contract clause extraction system, partnering closely with legal/procurement SMEs to create schemas, labeled datasets, and evaluation pipelines that improved accuracy from 70% to 85%. Also has experience architecting real-time voice/LLM systems with streaming microservices (Kafka, Kubernetes, gRPC/WebSockets) and an avatar chatbot pipeline (TalkingHead, Google TTS, AnythingLLM).

Experience

Education

Key Strengths

• Built production-oriented LLM app with batching, caching, async concurrency, and event-driven design (reduced latency ~10s to ~5–6s)
• Designs low-latency real-time voice + LLM systems using streaming, microservices, queues, and autoscaling (Kafka, Kubernetes, gRPC/WebSockets)
• Strong SME-driven product/ML iteration: translated legal expert workflows into schemas, labeled data, prompts, and evaluation metrics (improved 70% to 85%)
• Balances latency vs reliability trade-offs with pragmatic hybrid strategies (streaming + parallel full processing)