- 🔬 Published IEEE Researcher: Authored research on Early Detection of Mental Health Trends Using AI/ML published in IEEE Xplore (ICDT-2025).
- 🤖 AI Engineering (Outlier AI): Engineered and refined high-quality model training and alignment datasets, stress-testing complex prompt reasoning, algorithmic efficiency, and code generation across frontier LLMs.
- 💳 Distributed Payment Engine: Built a high-throughput transaction engine with ACID double-entry ledgers, Kafka event outbox, and Redis idempotency (k6 load-tested to 63 req/s at 0.00% error rate).
- 🛡️ Agentic AI Evaluation Platform: Engineered a production LangGraph multi-agent evaluation platform with pgvector RAG, Presidio guardrails, and real-time SSE observability.
Live Console: agentic-ai-eval-platform.vercel.app | API Docs: agentic-ai-eval-platform.onrender.com/docs
- Multi-Agent StateGraph: LangGraph engine coordinating Planner, RAG, Answer Synthesizer, and Evaluator nodes.
- Deterministic Guardrails: Microsoft Presidio PII redaction and heuristic prompt-injection scanners, short-circuiting malicious queries pre-LLM.
- Sub-10ms Semantic Caching: Redis dense vector similarity caching alongside PostgreSQL
pgvectorHNSW cosine search. - Real-Time Telemetry: Groundedness evaluation harness (0.88/1.00 average), human review triage escalation, and live SSE event streaming.
- ACID Double-Entry Ledger: Built a PostgreSQL ledger guaranteeing auditable transaction histories with strict data integrity.
- Idempotency & Fraud Mitigation: Implemented Redis idempotency keys for concurrent retry safety and sliding-window fraud detection filters.
- Event-Driven Settlement: Decoupled transaction settlement via an Apache Kafka transactional outbox pattern with dead-letter queue retries.
- Resilience Verified: Sustained 63 req/s under k6 load tests with 1.17s p95 latency and 0.00% error rate across 7,500+ transactions.
