Now — RAG & search at Stuvio Digital

Rahul Talepa, AI Engineer

AI Engineer — LLM systems

I build LLM systems that retrieve precisely, reason in the open, and only act when code says they may.

Résumé
Portrait of Rahul Talepa

Rahul Talepa

AI Engineer · Mumbai

2026
$ graph.stream("atelier.investigation_graph")
  1. ▍running…
0/6 nodes● streaming
Focus
RAG · Agents · Retrieval
Currently
Stuvio Digital
(01)About

AI Engineer
Mumbai, India

AI Engineer building production-grade LLM applications and RAG pipelines with LangChain and LangGraph. I work across semantic search, vector databases, embeddings, prompt engineering and tool calling. I've also built multimodal computer-vision search and multi-agent systems, and shipped them as scalable services on FastAPI and Docker.

I'mRahul,anAIengineerwhoturnslanguagemodelsintosystemspeoplecantrust.Ibuildretrievalthatfindstherightcontext,agentsthatshowtheirreasoning,andguardrailsthatdecidewhenthey'reallowedtoact.

Lakhs

of records served by production RAG retrieval

0+

automated tests guarding an agentic finance workflow

0

LangGraph graphs orchestrating a multi-agent research desk

0.00

CGPA, B.Tech Computer Science

How I build

Four rules I follow in every system I ship, from production RAG to the projects I build on my own time.

  1. 01

    LLMs recommend. Code decides.

    Models diagnose and propose; a deterministic policy layer approves, rejects or escalates before anything touches the real world.

  2. 02

    Retrieval is the product.

    Hybrid search, metadata filtering and retrieval evaluation matter more than the prompt. Good answers start with the right chunk.

  3. 03

    Every claim needs a citation.

    Citation validators reject invented references. If a source can't be traced, it doesn't reach the user.

  4. 04

    Degrade gracefully.

    Swappable providers, automatic fallbacks and rule-based paths keep the system useful when an API key or a model doesn't cooperate.

(02)Experience

Where the models meet production.

CurrentFeb 2026 — Present

Stuvio Digital

Junior AI Engineer

India

  • LangChain
  • RAG
  • Vector DBs
  • ViT
  • FastAPI
  • Docker
  1. 01

    Engineered production-grade LLM applications and scalable RAG pipelines using LangChain, semantic search, vector databases, embeddings, prompt engineering and tool calling, delivering accurate retrieval across lakhs of records.

  2. 02

    Developed an AI-powered multimodal image search system using Transformer-based computer vision, vector similarity search, metadata filtering and retrieval optimization.

  3. 03

    Built automated data ingestion, preprocessing, metadata enrichment, indexing and REST API integration pipelines with Python, FastAPI and Docker.

  4. 04

    Collaborated with backend teams to deploy scalable AI services to production.

Education · 2021 — 2025

B.Tech, Computer Science Engineering

D Y Patil University, Pune, India

8.33

CGPA

(03)Selected work

Systems that retrieve, reason & refuse.

Three projects I designed and built on my own, each testing a different way to make LLM output dependable.

01 / 03Agentic B2B invoice recovery

AI Revenue Recovery Agent

An auditable agentic workflow that detects overdue invoices, scores recovery probability, and lets an LLM recommend the next move, which a deterministic policy engine must approve before any action runs.

  • LangGraph + Gemini recommend actions; a policy engine enforces reminder caps, cooldowns and forced escalation
  • FastAPI + PostgreSQL backend with an XGBoost risk-scoring model, Kafka event streaming and Redis idempotency locks
  • LangGraph
  • Gemini
  • FastAPI
  • PostgreSQL
View on GitHub
120+
automated tests
02 / 03Multi-agent research intelligence

Atelier

A research desk where a Director agent decomposes technical questions into investigation plans, specialists gather evidence in parallel, and a synthesizer returns a cited report that won't pass hallucinated references.

  • LangGraph pipeline: Director → Specialists → Evidence Analyst → Synthesizer
  • Human-in-the-loop via interrupt(), with session state checkpointed in PostgreSQL
  • LangGraph
  • Gemini
  • FastAPI
  • PostgreSQL
View on GitHub
3
checkpointed graphs
03 / 03Hallucination-resistant RAG

Grounded Scripture Assistant

A Bible-grounded chat assistant that retrieves verses before answering, validates every reference, detects famous misattributions and returns citations with a confidence level.

  • ChromaDB + sentence-transformer embeddings over the World English Bible
  • Citation and verse-reference validation, plus misattribution detection for fake quotes
  • Gemini 2.5 Flash
  • ChromaDB
  • Sentence Transformers
  • FastAPI
View on GitHub
3
evaluation suites

The lab

Earlier ML & data work

All repositories
(04)Stack

The toolkit behind the systems.

Move over the field on the left. Every frame it runs a nearest-neighbour lookup, the same idea behind semantic search.

embedding space · k = 6
query nearest2D projection

Generative AI & Agents

Orchestration, reasoning and control

10
  • LangChain
  • LangGraph
  • AI Agents
  • Multi-agent systems
  • RAG
  • Prompt engineering
  • Tool / function calling
  • Structured output
  • MCP
  • Human-in-the-loop

Retrieval systems

Getting the right context, fast

08
  • Semantic search
  • Vector databases
  • Embedding models
  • Hybrid search
  • Metadata filtering
  • Retrieval evaluation
  • Document processing
  • ChromaDB

Backend & infra

Shipping it as a service

10
  • Python
  • FastAPI
  • REST APIs
  • SSE
  • PostgreSQL
  • MongoDB
  • Redis
  • Kafka
  • Docker
  • GitHub Actions

ML & vision

Models beneath the agents

07
  • Transformers
  • Vision Transformers
  • Image similarity
  • XGBoost
  • scikit-learn
  • Random Forest
  • Feature engineering

Languages & tools

Daily drivers

08
  • Python
  • SQL
  • TypeScript
  • JavaScript
  • Next.js
  • Git
  • Linux
  • Jupyter
(05)Beyond the code

Published, certified & printed.

A peer-reviewed paper, a generative-AI certification, and one book that has nothing to do with engineering.

Swipe

Peer-reviewed publicationMay 2025

AI Driven Drugs Traceability System

A published paper on applying AI to pharmaceutical traceability — authenticating medicine and tracking it across the supply chain so counterfeits can be caught before they reach a patient.

Journal
TIJER — International Research Journal
Issue
Volume 12, Issue 5 · May 2025
Paper ID
TIJER2505048
Impact factor
8.57 (Google Scholar)

Co-authors: Aniket Laxman Patil, Kunal Krishna Paste

Read the paperCertificateLinkedIn post
Certification

Generative AI Applications Specialist

Authorized by IBM

RAG with LangChain end to end: loading documents from varied sources, text-splitting strategies, configuring vector databases for embeddings, and building a QA bot on top of LLMs.

  • RAG
  • LangChain
  • Vector databases
  • Embeddings
  • QA bots
Verify on Credly
Cover of Subh Ratri by Rahul Talepa
Published authorOff the clock · my first book

Subh Ratri

A collection of late-night poems and reflections in simple Hinglish — love, loss, and the people worth staying up for. Not engineering. Still the thing I'm proudest of shipping.

Get a copyRead the post
(06)Contact

Let's build
something useful.

Hiring for LLM, RAG or agentic AI work? Or building something where retrieval and guardrails have to be right? My inbox is open.

Write to me