Utkarsh Gupta profile photo 1 on Tal

Utkarsh Gupta

Immediate Joiner | Sr. AI Engineer @ Trajector | Applied AI • LangGraph • LangChain • NLP • RAG • Python • AWS • GenAI • LLMs

Greater Bengaluru Area

5+ years of work experience

Trajector
Jul 20251 y 3 moPresent

Bengaluru, Karnataka, India

1) Cut inference cost by 60% and processing time by 80% on the production Python/FastAPI document-extraction pipeline that turns client documents (OCR or native text) into Pydantic-validated JSON for three lines of business, via per-document model routing, Bedrock prompt caching, and a zero-LLM fast path. 2) Run that pipeline in production at ~171K files and 1.4M pages/month for $0.24 per file, holding a 99.87% processing SLA with per-file cost and latency traced in Langfuse and releases shipped through GitHub Actions CI/CD. 3) Owned end-to-end an LLM call-information extraction pipeline (Claude Sonnet 4.6, AWS Bedrock, Snowflake) that populates the clinical intake questionnaire from nurse-led call transcripts, freeing high-cost clinical staff from live manual data entry and cutting call entry time by an estimated 40-50%. Extended it into PDIS, a proactive service that requests missing client documents. 4) Built and owned the LLM-as-a-judge evaluation framework, qualifying extraction, scoring a questionnaire against a human-reviewed golden dataset, benchmarking four extraction architectures, and selecting the one at 94.9% precision and 83.7% recall against a manual baseline of 84.4% and 62.2%. 5) Own the AI Draft Creator that assembles client data into filing-ready disability-claim documents: 54 prompt-template domains, PydanticAI orchestration, and model routing across AWS Bedrock (Claude Sonnet 4.5) and Azure GPT-4o, delivering ~27K drafts across two business lines.
Trajector
Jul 20241 yr 1 moJul 2025

Gurugram, Haryana, India

1) Redesigned lead intake system architecture, migrating from SQS queues to a MongoDB-driven model, reducing query latency by ~40% 2) Headed development for a workflow orchestration engine (frontend & backend), similar to n8n, to integrate e-sign, form-gen, questionnaires, faxing, document management, and lead management services for seamless automation. 3) Designed a form-gen-service to annotate and render PDFs in-browser, eliminating Adobe Acrobat licensing andsaving the company ~$12K annually, while reducing ops processing time by ~15+ minutes per intake packet. 4) Secured rank 2 out of 38 teams in an internal AI hackathon, leveraging open-source LLMs on AWS to automate document classification and packet verification with >90% accuracy. 5) Led POC for offline intake app using OPFS, enabling remote workflows with zero data loss and supporting field agents in low-connectivity regions. 6) Developed a multi-tenant architecture with AWS Cognito & Cerbos RBAC, with granular role permissions.
Medikabazaar
Oct 20239 mosJun 2024

Gurugram, Haryana, India

Project Name: API Drug 1) Led the development of the frontend architecture and integration with the AWS S3 bucket. 2) Managed the deployment and code release process of the front end, ensuring seamless updates and minimizing downtime for users. 3) Successfully maintained a 99.9% uptime rate throughout the project lifecycle. Project Name: Marketplace 1) Improved web performance KPIs like FCP, LCP, and CLS by 150% for the Medikabazaar shop. 2) Integrated New Relic for real-time logging/ monitoring of web performance issues to identify areas of improvement. 3) Spearheaded the development of the KYC module on the marketplace for frontend which resulted in a drop of 60% in order cancellation rates and assisted in a sanity check of genuine customers. 4) Implemented automation E2E testing with Cypress to mitigate production bugs and shorten the QA cycle. 5) Introduced structured data, dynamic sitemaps like brands, products, etc., and canonical URLs for optimized SEO. 6) Integrated Mixpanel to track the user journey funnels for product discovery, adding products to the cart, checkout, payment, and user drop-offs. 7) Revamped the UI of the PDP, listing page, homepage, and account page to create a more seamless and simplified UX.
Inspire Brands
Dec 20202 yrs 11 mosOct 2023

Hyderabad, Telangana, India

1) Built an IDP system for automating web apps for the company's US market (1Mn+ DAU) ensuring seamless UI/UX. 2) Achieved a 40% increase in the product backlog completion & refined user stories in collaboration with Product/UX. 3) Developed features that resulted in a 30% increase in user engagement, with NextJS, NodeJS, and REST services for brands (Arby's, Buffalo Wild Wings & Sonic). 4) Achieved a 15% reduction in QA bug reports by conducting thorough peer code reviews and implementing TDD.
GitHub
@utkarsh2709
LeetCode
@utkarsh2709

Utkarsh’s AI Stack

anthropic apis
anthropic apis
antigravity
antigravity
claude code
claude code
claude
claude
cursor
cursor
github copilot
github copilot

Education

Jss Academy of Technical Education

Bachelor of Technology - BTech, Electrical Engineering

Jss Academy of Technical Education

Join Utkarsh and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Utkarsh Gupta on Tal

Headline

Immediate Joiner | Sr. AI Engineer @ Trajector | Applied AI • LangGraph • LangChain • NLP • RAG • Python • AWS • GenAI • LLMs

About

Senior AI Engineer with 6+ years of software engineering experience, running production GenAI pipelines at ~171K files/month with measured cost, latency, and quality outcomes against human baselines. Builds LLM orchestration and evaluation frameworks, retrieval and multi-agent systems, and the scalable backend platforms that support them: AWS Bedrock, PydanticAI, Azure OpenAI, microservices, distributed systems, and payment platforms.

Location

Greater Bengaluru Area

AI stack

anthropic apis

antigravity

claude code

claude

cursor

github copilot

Work experience

Senior AI Engineer at Trajector, Jul 2025 - Present · 1 y 3 mo, Bengaluru, Karnataka, India, 1) Cut inference cost by 60% and processing time by 80% on the production Python/FastAPI document-extraction pipeline that turns client documents (OCR or native text) into Pydantic-validated JSON for three lines of business, via per-document model routing, Bedrock prompt caching, and a zero-LLM fast path. 2) Run that pipeline in production at ~171K files and 1.4M pages/month for $0.24 per file, holding a 99.87% processing SLA with per-file cost and latency traced in Langfuse and releases shipped through GitHub Actions CI/CD. 3) Owned end-to-end an LLM call-information extraction pipeline (Claude Sonnet 4.6, AWS Bedrock, Snowflake) that populates the clinical intake questionnaire from nurse-led call transcripts, freeing high-cost clinical staff from live manual data entry and cutting call entry time by an estimated 40-50%. Extended it into PDIS, a proactive service that requests missing client documents. 4) Built and owned the LLM-as-a-judge evaluation framework, qualifying extraction, scoring a questionnaire against a human-reviewed golden dataset, benchmarking four extraction architectures, and selecting the one at 94.9% precision and 83.7% recall against a manual baseline of 84.4% and 62.2%. 5) Own the AI Draft Creator that assembles client data into filing-ready disability-claim documents: 54 prompt-template domains, PydanticAI orchestration, and model routing across AWS Bedrock (Claude Sonnet 4.5) and Azure GPT-4o, delivering ~27K drafts across two business lines.

Senior Software Engineer (SDE-3) at Trajector, Jul 2024 - Jul 2025 · 1 yr 1 mo, Gurugram, Haryana, India, 1) Redesigned lead intake system architecture, migrating from SQS queues to a MongoDB-driven model, reducing query latency by ~40% 2) Headed development for a workflow orchestration engine (frontend & backend), similar to n8n, to integrate e-sign, form-gen, questionnaires, faxing, document management, and lead management services for seamless automation. 3) Designed a form-gen-service to annotate and render PDFs in-browser, eliminating Adobe Acrobat licensing andsaving the company ~$12K annually, while reducing ops processing time by ~15+ minutes per intake packet. 4) Secured rank 2 out of 38 teams in an internal AI hackathon, leveraging open-source LLMs on AWS to automate document classification and packet verification with >90% accuracy. 5) Led POC for offline intake app using OPFS, enabling remote workflows with zero data loss and supporting field agents in low-connectivity regions. 6) Developed a multi-tenant architecture with AWS Cognito & Cerbos RBAC, with granular role permissions.

SDE 2 - Frontend at Medikabazaar, Oct 2023 - Jun 2024 · 9 mos, Gurugram, Haryana, India, Project Name: API Drug 1) Led the development of the frontend architecture and integration with the AWS S3 bucket. 2) Managed the deployment and code release process of the front end, ensuring seamless updates and minimizing downtime for users. 3) Successfully maintained a 99.9% uptime rate throughout the project lifecycle. Project Name: Marketplace 1) Improved web performance KPIs like FCP, LCP, and CLS by 150% for the Medikabazaar shop. 2) Integrated New Relic for real-time logging/ monitoring of web performance issues to identify areas of improvement. 3) Spearheaded the development of the KYC module on the marketplace for frontend which resulted in a drop of 60% in order cancellation rates and assisted in a sanity check of genuine customers. 4) Implemented automation E2E testing with Cypress to mitigate production bugs and shorten the QA cycle. 5) Introduced structured data, dynamic sitemaps like brands, products, etc., and canonical URLs for optimized SEO. 6) Integrated Mixpanel to track the user journey funnels for product discovery, adding products to the cart, checkout, payment, and user drop-offs. 7) Revamped the UI of the PDP, listing page, homepage, and account page to create a more seamless and simplified UX.

Software Engineer - Fullstack at Inspire Brands, Dec 2020 - Oct 2023 · 2 yrs 11 mos, Hyderabad, Telangana, India, 1) Built an IDP system for automating web apps for the company's US market (1Mn+ DAU) ensuring seamless UI/UX. 2) Achieved a 40% increase in the product backlog completion & refined user stories in collaboration with Product/UX. 3) Developed features that resulted in a 30% increase in user engagement, with NextJS, NodeJS, and REST services for brands (Arby's, Buffalo Wild Wings & Sonic). 4) Achieved a 15% reduction in QA bug reports by conducting thorough peer code reviews and implementing TDD.

Education

Bachelor of Technology - BTech, Electrical Engineering, Jss Academy of Technical Education

Public links

github: https://github.com/utkarsh2709

leetcode: https://leetcode.com/utkarsh2709