> hello, world

Praveen Lal T H

|

4+ years building production GenAI at scale

praveen@deloitte ~ impact.sh
>
>_

about

I build production-grade GenAI systems that reduce manual work at scale. At Deloitte, I architected an enterprise contract intelligence platform using RAG pipelines and multi-agent workflows—cutting legal review time by 65%. 4+ years shipping intelligent systems: LLMs (Google ADK, LangChain, LlamaIndex, LangGraph), semantic retrieval, and full-stack GCP infrastructure.

4+
Years Experience
Deloitte
Company
4
Certifications
3
Projects
>_

skills

# Generative AI & LLMs
langchainllamaindexlanggraphvertex-airagfaisspineconeprompt-engineeringstructured-outputs
# Agents & Backend
multi-agentgoogle-adka2a-orchestrationfastapidjangomicroservicesevent-driven
# Data & Cloud
bigquerycloud-runcloud-functionspub-subpostgresqlelasticsearchcloud-storage
# MLOps & DevOps
dockerkubernetesci-cdmodel-monitoringdrift-detectiona-b-testingcloud-iam
# Languages
pythonsqljavascriptnode.jsreact
>_

experience

Deloitte Consulting India Pvt Ltd

AI Engineer

Jan 2022 – Present
Bengaluru, India
  • Architected production-grade GenAI platform using Vertex AI, LangChain RAG, and FAISS for contract intelligence (SoW, Order Forms, Amendments); reduced manual review time by 65%.
  • Designed scalable ingestion pipeline (Google Drive → Cloud Storage → Embeddings + BigQuery) processing 2,500+ contracts/month with 95% automated metadata accuracy.
  • Built hybrid retrieval engine combining semantic RAG and direct SQL on BigQuery, improving answer precision by 42%.
  • Engineered modular A2A architecture using Google ADK and async Pub/Sub event queues; cut query latency from 1.8s to 0.9s.
  • Developed multi-stage contract drafting pipeline using chained LLM prompts to extract clauses, validate business rules, and populate standardized templates.
  • Extracted and unified contract metadata in BigQuery, enabling structured reporting and 5x faster audit cycle.
  • Built library of reusable agent skills standardizing AI outputs across delivery teams, cutting ramp-up time by 50%.
  • Created Django dashboard supporting 10K+ monthly queries.
>_

projects

Deloitte

Contract Intelligence Platform

Enterprise GenAI platform for SoW, Order Form, and Amendment intelligence. Reduced manual review by 65% across legal teams with structured output validation.

Vertex AILangChainFAISSBigQueryCloud RunDjango
Personal

Gulabi Pani

A full-stack menstrual cycle tracking Progressive Web App (PWA) built for privacy-first, invite-only use. Features AI-assisted period prediction using weighted moving averages, daily symptom and mood logging, a 'Partner Share' read-only token link, admin-controlled user approval flow, email release note notifications, and a rich 'Aurora Glassmorphism' dark-mode UI. Built with Next.js 15 App Router, Prisma (PostgreSQL), NextAuth v5, and Resend for transactional emails.

Cycle prediction engine using weighted moving averages with configurable luteal phase

Daily log: symptom tracking, mood, flow intensity, pain scale

Partner sharing via shareable read-only token links

Next.jsTypeScriptPrismaPostgreSQLNextAuthResendVercelPWA
Visit site
Personal

MileagePro

A mobile-first fuel tracking app for bikes that calculates real-world mileage efficiency (km/L) using the start-fill method. Features automatic mileage calculation per fill-up, interactive charts for fuel spending and consumption trends, live Indian petrol/diesel prices per city via RapidAPI (cached for 12 hours), dark/light mode support, Google OAuth login, and an admin dashboard for viewing all users' statistics and fuel economy insights.

Live fuel price lookup for Indian cities with 12-hour DB caching

Automatic mileage calculation per fill-up using start-fill method

Per-user data isolation with role-based admin access

Next.jsTypeScriptNextAuthPrismaPostgreSQLChart.jsRapidAPI
Visit site
>_

certifications

AWS Certified Developer – Associate

Amazon Web Services

2025

Deep Learning Specialization

Coursera / deeplearning.ai

2023

CS50's Introduction to Computer Science

HarvardX

2022

Python for Everybody

Coursera / University of Michigan

2021

>_

contact

Open to AI Engineer roles. Let's build something intelligent.