Credentials
-
M.Sc. Computer Science (Data Analytics)
IIITM-K, 2020
-
B.Sc. Computer Science, Mathematics and Statistics
Kristu Jayanti College, 2017
Senior Engineer, Data Science at Reflections Info Systems. Based in Kerala, India.
I'm Arnol. I take generative AI from discovery to on-premise deployment.
I'm a forward deployed AI engineer with 5+ years in data science, NLP and computer vision. Most recently, in a team of three, I took on most of the technical delivery of an air-gapped English/Arabic assistant for a government client, including two months on site.
From discovery to handover
Selected work
Air-gapped bilingual AI assistant for a government client
A bilingual English/Arabic generative AI assistant, deployed on a government client's on-premise, air-gapped servers. Working in a team of three, I took on most of the technical delivery, from discovery sessions and GPU benchmarks to the on-site release deployments and the handover for user acceptance testing.
- Discovery
- Requirements
- Design
- Build
- Air-gap
- Deploy
- UAT handover
Deployed on-premise and handed over for user acceptance testing in October 2026.
Read the case studyOn-Prem Enterprise AI Platform
Self-hostable enterprise AI platform orchestrating LLM tool-calling across a federation of MCP servers, backed by a fully self-hosted inference stack.
Hybrid RAG Sales Intelligence
Proposal search assistant fusing vector search, graph traversal, and Reciprocal Rank Fusion behind a layered security pipeline.
AI-Powered Outreach Automation
Agentic LLM platform orchestrating multi-phase target discovery, data enrichment, fitness scoring, and branded document generation via MCP server architecture.
What I reach for
Toolbox
Client delivery
- Discovery
- Requirements (BRD)
- Solution design (SDD, LLD)
- On-premise and air-gapped deployment
- UAT support
LLM serving and infrastructure
- vLLM
- Tensor parallelism
- FP8 KV-cache quantization
- Load testing (Locust)
- Docker
- Kubernetes
RAG and agents
- Retrieval-augmented generation (RAG)
- Hybrid search (BM25, SPLADE)
- Cross-encoder reranking
- Qdrant
- Multi-agent systems
- Model Context Protocol (MCP)
Safety, governance and observability
- LLM guardrails
- Prompt-injection defense
- PII detection
- Qwen3Guard
- ISO 42001
Machine learning and computer vision
- PyTorch
- Transformers
- Fine-tuning (LoRA, QLoRA)
Engineering
- Python
- TypeScript
- FastAPI
- PostgreSQL
Things I've written up
Writing
-
10 min read
How I load-test self-hosted model serving before an air-gapped rollout
The method I use to benchmark speech, embedding and LLM serving on one GPU before it ships to an air-gapped site: the test bench, mixed-load and soak runs, a text-to-speech fix, an FP8 check, and what changed once I was on site.
-
9 min read
I Stopped Using Basic RAG in 2026. Here's What Replaced It.
Basic RAG is a one-trick pony. I built a hybrid system that fuses vector search, graph databases, and intelligent routing - with production security. Here's the architecture, explained simply.
-
9 min read
Building a Multi-Agent Log Analytics Platform From Scratch
How we designed a 4-agent event-driven pipeline for high-volume logs, with semantic error grouping, anomaly detection, and intelligent workflow routing.
Where I've worked
Experience
-
Aug 2025 to Present
Senior Engineer - Data Science
Reflections Info Systems Pvt. Ltd.
I build and deliver generative AI systems, for clients and for the firm itself.
- For a government client, working in a team of three, I took on most of the technical delivery of an air-gapped bilingual English/Arabic RAG assistant. I ran the discovery sessions, co-wrote the requirements and design documents, and converted the platform to run fully offline.
- I configured the self-hosted model serving with vLLM, benchmarked it under burst load, with every request succeeding, and re-tuned the Arabic model for more than a quarter more output throughput. Working on site in October 2026, I ran multiple release cycles on the client's servers in a short timeline, turning acceptance-test findings into same-week fixes before the UAT handover.
- I rebuilt SalesBot, the firm's RAG assistant for sales proposals, routing each query to vector, keyword or graph search (Qdrant, BM25, Neo4j) behind layered security, and shipped it to UAT and internal production.
-
Dec 2024 to Jul 2025
Senior Data Scientist - Consultant
Digital University Kerala
Led computer vision research identifying individual cattle from muzzle images, validated across 197 animals. Co-authored two research manuscripts: handwritten text recognition (submitted to IJDAR) and a citation-network study of agricultural robotics.
-
Jun 2024 to Nov 2024
Senior Software Engineer - AI/ML
Techversant Infotech
Built RAG applications with memory for multi-turn conversations, face recognition systems and YOLO-based proctoring tools for enterprise products.
-
Sep 2023 to Jun 2024
Senior Engineer - Data Science
Digital University Kerala
Led a team of three building a multilingual knowledge-management platform for a national space research centre: OCR ingestion, cross-lingual hybrid search, and RAG answers that cite their source pages.
-
Jun 2021 to Aug 2023
Data Analyst
Digital University Kerala
Built automated data pipelines and reporting visualizations, and optimized database performance.
-
Sep 2019 to Sep 2020
Research Fellow
ICFOSS
Built a Malayalam morphological analyzer and sentiment analysis systems for Indian languages, and researched YouTube comment data.
Say hello
Get in touch
Email is the fastest way to reach me.