Headline

AI Engineer • Lead Architect @ IRIS • Specializing in LLMOps, High-Performance RAG & Agentic Workflows (LangGraph/Kubernetes)

About

/ AI Engineer building production Generative AI, LLM, RAG, Agentic AI and AI Platform systems. / At Sylox, I build IRIS, working across multi-stage AI detection, LangGraph workflows, cloud integrations, secure multi-tenancy, FastAPI/PostgreSQL and production reliability. / Built Cerevra, a deterministic RAG engine with crash recovery and namespace isolation -- sub-50ms p99 retrieval at 500+ QPS, backed by 2,000+ resilience and chaos tests. / Previously built production legal-tech RAG with hybrid retrieval, reranking and matter-level isolation. Also shipped 11 merged open-source PRs across 7 projects, including Unsloth, VictoriaMetrics and Prefect. / Python • FastAPI • LangGraph • RAG • LLMOps • pgvector • PyTorch • Kubernetes • Kafka :: I build AI systems that are reliable, secure and boring in production.

Location

Pune District

AI stack

antigravity

claude code

codex

notebooklm

ollama

Pinecone

Work experience

AI Engineer at Sylox, Feb 2026 - Present · 7 mos, Pune District, Maharashtra, India, \ Architected the core of IRIS, a multi-tenant DSPM platform, shipping 44 FastAPI routers and 400+ async REST endpoints across an 11-schema Postgres backend. \ Optimized backend performance for production scale, sustaining p95 latency under 500ms at 80-concurrency autoscale across 1M+ records. \ Engineered a 7-stage LangGraph pipeline for India-specific PII discovery using Presidio, spaCy NER, Claude-based verification, and custom validators, improving detection accuracy by 18%. \ Built 105 enterprise-grade cloud and data-source connectors, including SAP HANA, Snowflake, and Databricks, for large-scale unstructured data ingestion and classification. \ Designed hybrid RAG infrastructure using pgvector and full-text search, functioning as an Azure AI Search-style retrieval layer for compliance-focused conversational agents. \ Hardened platform security with OIDC/OAuth2, SCIM 2.0, per-endpoint RBAC, and cross-tenant isolation for DPDP Act 2025 compliance. \ Owned CI/CD on GCP Cloud Build with 30-gate verification, 161 idempotent Alembic migrations, and 80–90% diff coverage.

AI Engineer at LOPHILS INC., Jun 2025 - Jan 2026 · 7 mo, Pune District, Maharashtra, India, \ Built and shipped a production Retrieval-Augmented Generation system with strict matter-level namespace isolation to prevent cross-case retrieval. \ Designed scalable ingestion pipelines with SHA-256 deduplication, rich metadata extraction, and 1GB+ large-file streaming support. \ Implemented Pinecone-based vector search using 1024-dimensional embeddings and query-side routing for deterministic retrieval. \ Developed an admin panel for namespace lifecycle management, including create, monitor, and delete workflows. \ Deployed end-to-end AI infrastructure without breaking existing n8n automation workflows.

Education

Bachelor of Engineering - BE, Information Technology, Savitribai Phule Pune University (SPPU)

Public links

github: https://github.com/wtfashwin

medium: https://medium.com/@wtfashwin

leetcode: https://leetcode.com/wtfashwin

codeforces: https://codeforces.com/profile/wtfashwin

youtube.com: https://youtube.com/@wtfashwin

wtfashwin.github.io: https://wtfashwin.github.io/Portfolio/