01MCP client
Atlas Co-Pilot
A domain-specific co-pilot that connects to Salesforce and Informatica MDM through several MCP servers and gives support teams contextual recommendations.
↳Cut support ticket volume by 40%
- MCP
- Salesforce
- Informatica MDM
I work inside your team, find where AI earns its place, and ship it to production.
I'm a Forward Deployed AI Engineer at Data Color AI. I work directly with customer teams to understand how they work, figure out what AI can realistically do for them, and then build it: LLM agents, MCP integrations, retrieval systems and AI steps inside their data pipelines. I work across AWS, GCP and Azure, and I ship the whole stack.
Samith Deshai Siddo
Forward Deployed AI Engineer · Dallas, TX
Forward deployed: close to the problem.
I don't start from a model. I start with the people doing the work, learn their systems and data, and find the places where AI changes the outcome. Then I build it with them and stay until it runs on its own.
- 01
Discover
Sit with the team
I learn how the work happens today: the systems, the data, the handoffs, and where people lose hours.
- 02
Scope
Find where AI fits
I separate what an LLM is actually good at from what a rule or a query should do, and agree on a number that defines success.
- 03
Build
Ship a working agent
I build a prototype on their real data within days, then harden it with evals, guardrails and human review where mistakes are costly.
- 04
Hand off
Leave it running
It goes live with monitoring, cost tracking, audit trails and documentation, so the team owns it after I step back.
Agents that work inside enterprise data.
Production systems I've built alongside customer teams. They connect to MDM platforms, ERPs, CRMs and data pipelines, and they bring in a person for the decisions that need one.
02Human-in-the-loop agent
AI Data Steward
When the same customer shows up twice in Reltio, a data steward has to decide whether to merge the records. This agent does the investigation for them. It compares the two records, checks what a merge would change and whether it breaks any data rules, then recommends merge, reject or wait. A person makes the final call, and every decision is recorded.
↳Stewards review a recommendation instead of researching every pair
- Azure OpenAI
- Reltio MCP
- Service Bus
- Cosmos DB
03Document agent
DocuAssist
An agent built on Informatica's AI Agent Engineering Platform. It turns unstructured documents into governed, structured data sources using RAG-based extraction and reasoning.
↳Made document ingestion 3× faster
- RAG
- Informatica
- Hugging Face
04LLM matching · post-acquisition
Smart Mapping
When a company is acquired, its service codes live in its own ERP and CRM and look nothing like the parent's catalog. Smart Mapping has an LLM do the mapping. For each code, Gemini reads the description, weighs it against the closest catalog entries, and decides which one it is. Search and rules only narrow the options first. When Gemini isn't confident, the code goes to an analyst.
↳20 acquired brands onboarded, about 70% of codes auto-accepted
- Gemini
- BigQuery VECTOR_SEARCH
- Cloud Run
- TF-IDF
05Data engineering
AI-Native Data Pipelines
I build LLMs into data engineering pipelines. On Databricks, a Spark validation framework has LLMs profile the data and propose quality rules, a person reviews them, and they run at scale. Other LLM steps classify and enrich records along the way. MLflow tracks the results.
↳Better data reliability and observability across pipelines
- Databricks
- Spark
- MLflow
- Hugging Face
Built on my own time, shipped to real users.
A prompt engineering product with paying subscribers, a voice agent that won a hackathon, and a desktop app that turns lectures into fact-checked courses.
Prompt Weaver
2025Turn napkin sketches into production prompts.
A prompt engineering platform that turns sketches, screenshots, voice and screen recordings into structured prompts for AI coding tools. It's live and has paid subscriptions.
- Three prompt modes, each tuned for a different kind of coding tool: general LLMs, Bolt/Lovable and Cursor/Windsurf
- Whiteboard drawing, screen-share animation capture, custom palettes and UI-to-ER diagram conversion
- Full subscription billing with usage tracking and plan enforcement through webhooks
- TypeScript
- React
- PostgreSQL
- LLMs
promptweaver.netlify.app
1st place, UNT GradInnoHack
MediCall
2025A voice agent that books your doctor's appointment.
A real-time voice AI receptionist. You call it and talk normally to book, reschedule or cancel medical appointments.
- Gemini for reasoning, Deepgram for speech-to-text and ElevenLabs for natural-sounding speech
- LangChain tool calling, with the tools running as serverless functions
- Appointments sync to Supabase in real time while the call is happening
- Voice AI
- LangChain
- React
- Supabase
Paste a lecture video or type a topic, pick your level and model, and generate. Your courses sit below.
LectureForge
2026Turn any lecture into a researched, cited course.
A desktop app that turns a study video into a mini-course with lessons, a quiz, flashcards and a tutor. It checks the lecture against the live web and flags anything that's out of date.
- Paste a video link or type a topic and get a full course: lessons at your level, a quiz and flashcards
- Fact-check badges on every claim. If the lecture is out of date, you see what it said, what's true now, and the sources
- Watch it work live in Mission Control as it plans, researches each concept and builds the course
- Ask the tutor anything. It answers from the course, searches the web, and draws charts and 3D scenes to explain
- Spaced-repetition review, a knowledge graph of what you've learned, one-click re-check of facts, and export to Notion
- Deep Agents
- Tavily
- MCP
- FastAPI
- React
- Tauri
How agents actually work, drawn out.
Illustrated explainers on the parts of AI engineering that tutorials skip: harnesses, MCP, tool calling, memory and context.
Where I've worked
Data Color AI
Forward Deployed AI Engineer
2025 — Now · Dallas, TX
- Work directly with customer teams to understand their workflows and data, find the use cases where AI pays off, and take them from prototype to production.
- Build agentic AI products for enterprise master data management, connected to MDM, ERP and CRM systems through MCP.
- Built custom MCP servers that connect our agents to enterprise platforms, exposing each system's data and actions as tools the agents can call safely.
- Build AI into data engineering pipelines on Databricks and Spark, including LLM-generated data quality rules, enrichment and MLflow monitoring.
- Built secure integrations between MCP servers and third-party clients with AWS Cognito OAuth and the Bedrock AgentCore runtime.
- Deploy open-source Hugging Face embedding and instruction-tuned models in RAG pipelines, tuning them for latency and retrieval quality.
Club InQuizitive
Full Stack Developer
2021 — 2023 · Hyderabad, India
- Led a team of developers building the club's main site and several event platforms from scratch.
- Shipped responsive React and Tailwind apps with PostgreSQL and Firebase backends for live event registration.
Wins, papers and launches
- 20251st place, UNT GradInnoHackWon for building MediCall
- 2025Google Gemini × Pipecat Hackathon at Y CombinatorVoice and agentic AI hackathon during SF Tech Week
- 2023IEEE publicationDesign and implementation of a product recommendation system
- 2020Flutter app on the Play Store500+ users, 4.5★ rating
What I build with, end to end.
From the model call to the queue it runs on. The highlighted tools are what I use every week. The rest I've shipped with.
Agents
01Loops, tools and guardrails
- LangGraph
- MCP
- Tool calling
- Human-in-the-loop
LangChain · Agent memory · Prompt caching · Voice agents
Retrieval
02Getting the right context in
- Hybrid RAG
- Vector search
- LLM re-ranking
TF-IDF · BigQuery VECTOR_SEARCH · Pinecone · Chroma · Embeddings
Models
03Where the reasoning runs
- AWS Bedrock
- Vertex AI · Gemini
- Azure OpenAI
Bedrock AgentCore · Claude · Hugging Face · Open-source LLMs
Data
04Pipelines AI plugs into
- Databricks
- Spark
- BigQuery
MLflow · PostgreSQL · Cosmos DB · MongoDB · Supabase
Cloud & infra
05Running it in production
- GCP Cloud Run
- Azure Service Bus
- Docker
Container Apps · Artifact Registry · GCS · AWS Cognito · OAuth
Product
06The rest of the stack
- Python
- TypeScript
- React
- FastAPI
Next.js · Node.js · Java · Flutter · pytest
Education & certifications
University of North Texas
2025M.S. Computer Science
GPA 3.90 · NLP, Information Retrieval, Software Development for AI
BV Raju Institute of Technology
2023B.Tech Computer Science
Data Structures, Operating Systems, Design Patterns
- MCP: Build Rich-Context AI Apps — Anthropic × DeepLearning.AI
- LangChain for LLM Application Development — DeepLearning.AI
- JavaScript RAG Web Apps with LlamaIndex — DeepLearning.AI