Nitin Manral

Senior AI Engineer at Collibra.

Building production AI systems — agentic architectures, RAG pipelines, and LLM evaluation frameworks. Based in London.

Projects
ModelMind

Text-to-SQL platform with dual-LLM architecture. Serves 500+ daily queries across 70+ countries with natural language database access via ChromaDB embeddings and Cloud Run.

Python ChromaDB OpenAI Cloud Run BigQuery
DeepAnalysis

8-agent autonomous research system built on Google ADK with supervisor-worker architecture. Features dual FAISS indexes, iterative refinement, and ReAct patterns for deep organisational analysis.

Google ADK Gemini 2.5 FAISS Multi-Agent ReAct
AllRound

Hybrid BM25 + semantic search knowledge base using Reciprocal Rank Fusion. Aggregates GitHub, Google Docs, Confluence, and other organisational sources into a unified retrieval layer.

BM25 RRF Fusion Semantic Search FastAPI
Formula E Race Intelligence

Full-stack race analytics dashboard built from 4,123 scraped CSV files across 12 seasons. React frontend with FastAPI backend for real-time race intelligence and historical analysis.

React TypeScript FastAPI Tailwind Vite
Meeting Assistant

macOS overlay app built in Swift using ScreenCaptureKit for audio capture and SFSpeechRecognizer for real-time transcription, piped through Claude API for live meeting intelligence.

Swift ScreenCaptureKit Claude API Xcode
Experience
Apr 2025 → Present
Senior / Staff AI Engineer
Collibra
London, UK
AI engineering for data intelligence platform.
Feb 2024 → Mar 2025
Staff Data & AI Engineer
Decathlon
London, UK
Built and shipped AI systems serving 70+ countries. Architected DeepAnalysis (8-agent research system), ModelMind (text-to-SQL, 500+ daily queries), and MCP server with 8 specialised tools. Created reusable patterns adopted by 50+ teams.
May 2023 → Jan 2024
Lead Data Engineer
Decathlon
London, UK
Embedded with 50+ data teams to deploy centralised DBT infrastructure. Drove 85% adoption through technical content, tutorials, and reusable components. Reduced data incidents by 60%.
Jan 2022 → May 2023
Data Engineer
Decathlon
London, UK
Built production data pipelines processing millions of records daily using Airflow, Databricks, and Spark. Designed MLOps workflows for model deployment and monitoring.
Sep 2021 → Nov 2021
Data Scientist Intern
StormHarvester
Belfast, UK
Automated data pipelines for sensor data from 100+ IoT sites. Built AWS integrations and interactive dashboards, reducing manual preprocessing by 70%.
Nov 2017 → Oct 2020
Senior System Engineer
Infosys
Bangalore, India
Led team of 10 engineers for German enterprise client. Built data integration applications and automated workflows, achieving 25% productivity gain.
Writing & Thinking
Context Management in Long-Running Agentic Workflows
Preventing context window explosion in multi-agent coordination
The Missing Prefrontal Cortex in Agent Architecture
Structured deliberation, situation classification, and inhibition layers
Why AllRound Failed: Building for Organisations vs Individuals
A technically successful product with an adoption failure