Skip to content
Krish Patel
AI EngineerSan Francisco Bay Area

I put AI agents into the systems enterprises already run on.

I build the integrations, MCP servers, and AWS serverless infrastructure that take an AI agent from demo to production, including Amazon Connect, Genesys, and Salesforce deployments running live with customers. I also ship open-source retrieval tooling.

4 days → 5 min
Demo integration setup, via a self-serve MCP platform
40+
Enterprise prospects scoped, prototyped, and handed off
~10 → ~4 days
Turnaround on new partner integrations
87–91%
Retrieval quality retained by isotrieve on BEIR

01/Experience

Shipping AI into production for enterprise customers.

AI Engineer (Pre-Sales) · eGain

Forward-deployed work between the customer and the product: I scope requirements on customer calls, build the integration, and hand it off to post-sales running.

Aug 2025 – Present
  • Integrated eGain's AI Agent for Contact Center with Amazon Connect, Genesys, and Genesys + Salesforce. The integrations run live across thousands of agents, and intent validation serves 20K+ requests a month. Co-developed a Gemini voice assistant in 30 languages and cut ~2.5s from each response.
  • Built a self-service integration platform on AWS Lambda + API Gateway with an MCP server and hundreds of tools (Salesforce, Jira, ServiceNow, Calendar, Maps). It has been used in 50+ demos and cut demo setup from 4 days to under 5 minutes.
  • Built offline mode for eGain's knowledge portals: a service worker with device-aware caching, SQS-queued actions, and encrypted IndexedDB storage. It is now being productized as its own product.
  • Delivered demos, PoCs, and integrations for 40+ enterprise prospects in financial services, credit unions, and travel. Worked with 7 technology partners and cut new-integration turnaround from ~10 to ~4 days.
  • Building an AI sales agent with memory across deals, pulling context from MS Dynamics, Teams (Microsoft Graph), and SharePoint. A reliability-weighted memory layer turns 5–7 minute lookups into seconds.
  • Shipped RFP automation (~3 RFPs a week for 6+ months) and 4 knowledge-sync connectors, and migrated the AI Agent to schema-validated Pydantic tool calling.
PythonTypeScriptAWS LambdaAPI GatewaySQSMCPVue.jsAmazon ConnectGenesysSalesforce
  1. Jun 2024 – Aug 2024
    Pleasanton, CA

    AI Intern · dotSolved Systems

    • Built Salesforce Einstein AI proofs of concept for customers, contributing to a 30% increase in client adoption.
    • Built fine-tuned LLM pipelines (Llama 3, Unsloth) on HubSpot APIs, with 60% faster email creation and 25% higher engagement.
    Salesforce EinsteinLlama 3UnslothHubSpot API
  2. Aug 2024 – Sep 2024
    Remote

    Software Engineering Intern · Ximple Solutions

    • Built sales-forecasting pipelines with Meta Prophet plus neural-net, regression, and KNN models, reaching 90% year-over-year accuracy.
    • Sped up Snowflake and Palantir Foundry workflows, cutting query time 40%, and integrated Keycloak role-based 2FA into Angular apps.
    ProphetSnowflakePalantir FoundryAngularKeycloak
  3. Mar 2023 – Oct 2025
    Los Angeles, CA

    DevOps Project Lead · UCLA Student Media

    • Led a 10-person team deploying the Apply platform on DigitalOcean + Django for 4,000+ applicants.
    • Made Apply 9× faster by caching and optimizing slow Django queries, which also sped up report generation.
    • Deployed Timesheet and Inventory apps behind Nginx with custom CI/CD, reducing load time 60% and latency 30%.
    DjangoPostgreSQLDigitalOceanNginxCI/CD

02/Projects

Open-source retrieval tooling, and models built from scratch.

isotrieve

Embedding-model migration for vector databases

Switching embedding models normally means re-embedding your entire corpus. isotrieve learns a mapping between the old and new embedding spaces from ~2K calibration texts, migrates the store in place, and gates every migration on measured retrieval retention.

$ pip install isotrieve
  • 87–91% retrieval quality retained on BEIR
  • Procrustes, Ridge, low-rank, residual-MLP, and contrastive mappings
  • Adapters for Chroma, pgvector, Qdrant, Pinecone, LangChain, LlamaIndex
  • CLI and 232+ passing tests
PythonNumPyPyTorchBEIRPyPI

03/Skills

The stack I work in day to day.

AI & Agents
MCP serversTool calling (Pydantic)Agentic workflowsRAGLangChainLlamaIndexPyTorchMLXLLM fine-tuning
Vector DBs
ChromapgvectorQdrantPinecone
Languages
PythonTypeScriptJavaScriptSQLC/C++JavaBash
Cloud & Infra
AWS LambdaAPI GatewaySQSEC2AzureDockerCI/CDNginxFirebase
Integrations
SalesforceMS DynamicsServiceNowJiraZendeskSharePointMicrosoft GraphGenesysAmazon ConnectCisco
Frameworks & Data
ReactNext.jsVue.jsNode.jsDjangoFlaskPandasMongoDBSnowflake

04/About

Most of my work happens between the customer and the product. I join the call, figure out what the system actually needs to talk to, build the integration, and stay on it until it runs in production. When something breaks, I fix it there.

I got into AI by building models from scratch, including transformers, Llama 3, and RL agents, because I wanted to know how they work. Today that matters most when a retrieval pipeline or tool-calling agent misbehaves in someone else's stack, and isotrieve came from the same instinct.

Outside work, I run an annual AI workshop for high-school students through exploretech.LA. I'm also a tennis fan, and I built match-analysis tools for UCLA's Division I team.

Education

University of California, Los Angeles

B.S. Computer Science · May 2025

Machine LearningDeep LearningNatural Language ProcessingComputer VisionOperating SystemsComputer ArchitectureAlgorithms
Leadership

External Director · exploretech.LA

Jul 2023 – Present

Secured $23K in sponsorships, grew engagement 260%, and run an annual AI workshop for 500+ high-school students.

05/Contact

Have a hard integration or an AI deployment to ship? Let's talk.