Skip to content
Vansh Singla — home

Open to Forward Deployed Engineer roles

Gurgaon, India

Vansh Singla — Forward Deployed Engineer I take Generative AI from idea to production — inside the customer’s own environment.

AI Python Engineer with 3 years embedded with enterprise clients as the sole engineer on each account — shipping RAG pipelines, voice AI, and agents that migrate and debug legacy 4GL code.

embedded with enterprise clients
3yrs
enterprise engagements, sole engineer
5
uptime at 10,000+ daily API requests
99.5%
critical outages since Oct 2023
0

An illustrative migration-agent console: the agent grounds itself in 4GL grammar references, annotated code samples and RAG over the client’s code, translates a 4GL module to .NET, compiles and tests it, then queues it for human review. Result: 17 legacy modules migrated to .NET.

IllustrativeThe compile-and-test loop behind the EPS 4GL → .NET migration

Embedded with

  • BCG
  • Sony LIV
  • Dentsu
  • Americana
  • EPS
Latest build Wholesale commerce & store-management platform

Impact, in numbers

  • ~60%

    faster proposal generation

    Internal presales tool · LangChain + GPT-4

  • 35%

    faster content APIs

    Sony LIV · 1M+ active users

  • 17

    legacy 4GL modules migrated to .NET

    EPS · code-migration agent (Claude API)

  • ~95%

    voice-order accuracy

    Americana · hundreds of live orders daily

  • 40%

    higher retrieval accuracy vs keyword baselines

    RAG agents · OpenAI embeddings + pgvector

  • Days to minutes

    enterprise report turnaround

    Dentsu · FastAPI + LLM narratives

  • 20–30

    production defects resolved by a bug-fixing agent I built

    EPS · in daily use by the maintenance team

  • 15+ hrs

    reclaimed every week for a presales team of 10+

    GenAI presales tool · Kellton

01 / About

Embedded with the customer, accountable for production.

I’m an AI Python Engineer who has spent the last 3 years embedded with enterprise clients — BCG, Sony LIV, Dentsu, Americana and EPS — as the sole engineer on each account.

I specialise in taking Generative AI from idea to production in the customer’s own environment: RAG pipelines, voice AI, and agents that migrate and debug legacy 4GL code.

The results: ~60% faster proposal generation, 35% faster APIs for a 1M+ user platform, enterprise reporting cut from days to minutes, and 17 legacy modules migrated to .NET — with zero critical outages across 3 years of production ownership.

At a glance

Portrait of Vansh Singla

Vansh Singla

Forward Deployed Engineer

Available — Open to Forward Deployed Engineer roles
Based in
Gurgaon, IndiaIST · UTC+5:30
Currently
AI Python Engineer / Software Developerat Kellton Tech Solutions Ltd
Focus
Generative AI, from idea to production
Latest build
Brijesh Poshak WalaWholesale commerce & store-management platform

What I bring to a forward deployed role

  • Embedded delivery

    Scope, build, demo and support production systems directly with client business and technical stakeholders, translating between executives and engineering teams.

  • Prototype-to-production speed

    Working demo in days (Streamlit / FastAPI), then hardened with CI/CD, monitoring and evaluation to 99.5% uptime — and handed over so the client can run it without me.

  • Applied LLM depth

    RAG, embeddings, agent orchestration and evaluation (Claude API, OpenAI API, LangChain, Copilot Studio), including agents operating in a language the model had never seen.

  • Customer-environment deployment

    Ship into client cloud tenants (GCP, Azure, AWS) with Docker and GitHub Actions, working within enterprise security, access and compliance constraints.

How an engagement runs

  1. Discover

    Requirement discovery with client business and technical stakeholders.

  2. Prototype

    A working demo in days — Streamlit or FastAPI — iterated through stakeholder demos.

  3. Harden

    CI/CD, monitoring, evaluation and guardrails, hardened to 99.5% uptime.

  4. Hand over

    User training and production handover, so the client can run it without me.

02 / Experience

Three years of production ownership.

Client-embedded delivery across 5 enterprise engagements — with zero critical outages since Oct 2023.

  1. to Present Current

    Kellton Tech Solutions Ltd

    AI Python Engineer / Software Developer

    Gurgaon, India

    Client engagementsEPS, Americana, Dentsu, Sony LIV, BCGSee the engagements
    daily API requests
    10,000+
    uptime
    99.5%
    daily client users
    50+
    critical outages
    0
    • Sustain 10,000+ daily API requests at 99.5% uptime across production FastAPI and Django services on GCP (GCS, Firebase) and Azure, with zero critical outages since Oct 2023.
    • Run requirement discovery, iteration demos, user training and production handover with client stakeholders; delivered tools are in daily use by 50+ client users across 5 organisations.
    • Built an internal Generative AI presales tool (LangChain, GPT-4) that drafts proposals from client briefs and past engagements, cutting proposal generation time ~60% and reclaiming 15+ hours/week for a presales team of 10+.
    • Architected RAG conversational agents (OpenAI embeddings + pgvector semantic search) with a retrieval-quality evaluation harness against keyword baselines, reaching 40% higher retrieval accuracy; added guardrails and latency/cost monitoring in production.
    • Halved deployment time and eliminated environment-drift bugs across staging and production by introducing GitHub Actions CI/CD pipelines and Docker-based builds.

    Stack

    • Python
    • FastAPI
    • Django
    • GCP
    • Azure
    • LangChain
    • OpenAI API
    • pgvector
    • Docker
    • GitHub Actions
  2. to

    The Sparks Foundation

    Data Science Intern

    Remote

    faster than manual auditing
    3x
    less manual audit effort
    ~40%
    • Built anomaly-detection models in Python and Scikit-learn that identified fraudulent records 3x faster than manual auditing.
    • Reduced manual audit effort ~40% and improved reporting reliability by designing an automated data-validation pipeline.

    Stack

    • Python
    • Scikit-learn
    • Anomaly detection

03 / Client engagements

Five enterprises. One engineer on each account.

Sole engineer on every account: scoping with business and technical stakeholders, building inside the client’s environment, and owning it in production.

  • EPS

    Bug-fix agent in daily use

    Agentic code migration

    Legacy 4GL modernisation with AI agents

    legacy modules migrated to .NET
    17
    production defects resolved
    20–30
    maintenance team uses it daily
    5-person

    The challenge

    The client relied on a legacy 4GL codebase that no modern LLM had training coverage of and that few engineers could maintain. I owned the AI approach end to end, from feasibility spike through production rollout.

    How it works

    Code migration: Legacy 4GL, then Grounding layer (grammar, samples, RAG), then Migration agent, then Compile & test (validation loop), then Human review, then .NET.

    Bug fixing: Defect report, then Locate fault, then Proposed patch + explanation.

    What I built

    • Built a code-migration agent that translated the 4GL application to .NET module by module, with automated compile-and-test validation loops so generated code was verified before human review; migrated 17 modules.
    • Built a bug-fixing agent that ingests defect reports, locates the fault in the legacy 4GL code, proposes a patch and explains the change; it has resolved 20–30 production defects and remains in daily use by the client’s 5-person maintenance team.
    • Taught the model 4GL from scratch by engineering a grounding layer: language grammar references, annotated code samples, and RAG over the client’s own codebase and documentation.
    • Claude API
    • AI agents
    • RAG
    • .NET
  • Americana

    Live in production

    Voice AI · Enterprise agent

    Enterprise AI agent & voice ordering system

    order accuracy on live voice orders
    ~95%
    of live orders processed daily
    Hundreds
    How it works

    Voice ordering: Spoken order, then Speech recognition, then LLM intent & item extraction, then Catalogue validation, then Structured order.

    What I built

    • Built and deployed an enterprise AI agent end to end on Copilot Studio and Azure inside the client’s own tenant, from use-case definition through production rollout.
    • Designed and delivered a voice ordering system that converts spoken customer orders into structured, validated transactions — speech recognition, LLM-based intent and item extraction, and validation against the client’s catalogue — now in production processing hundreds of live orders daily at ~95% order accuracy.
    • Copilot Studio
    • Azure
    • Voice AI
    • LLM extraction
  • Dentsu

    GenAI reporting

    GenAI reporting microservices & KPI dashboards

    report turnaround
    Days → minutes
    • Cut enterprise report turnaround from days to minutes by delivering FastAPI microservices with integrated LLM features for summarisation and narrative generation.
    • Built real-time React.js and Streamlit KPI dashboards on Google Cloud, giving client stakeholders live visibility they previously assembled by hand.
    • FastAPI
    • GCP
    • LLMs
    • React.js
    • Streamlit
  • Sony LIV

    Backend performance

    Streaming backend performance

    faster content API responses
    35%
    active users served
    1M+
    • Accelerated content API response times 35% for a platform serving 1M+ active users by optimising the media-ingestion backend and tuning PostgreSQL/MySQL query performance.
    • Python
    • PostgreSQL
    • MySQL
  • BCG

    Analytics automation

    Reporting & analytics automation

    of manual work removed per reporting cycle
    2–3 days
    • Removed 2–3 days of manual data processing from every reporting cycle by automating end-to-end reporting pipelines and analytics dashboards for consulting teams.
    • Pandas
    • NumPy
    • Matplotlib

04 / Projects

Things I’ve built.

From a full wholesale commerce platform to the GenAI tools behind faster proposals and better retrieval.

Client build · 2026

Brijesh Poshak Wala

Wholesale commerce & store-management platform

A complete wholesale platform for a pooja-goods business: a bilingual (English / हिंदी) storefront, Android & iOS apps with a staff dashboard, and an in-store QR-code POS — all running on one shared PostgreSQL inventory.

web pages across storefront & admin
65+
automated backend tests
125+
Playwright end-to-end tests
18
languages — English & हिंदी
2

Illustration: the bilingual storefront in a browser, a phone running the in-store QR-code POS just after a pack is scanned, and a GSTR-3B report marked ready.

What I built

  • Bilingual storefront

    English / हिंदी with one switch. Product and category texts left blank in Hindi are translated automatically by AI.

  • Wholesale access control

    Mobile-number OTP sign-in, owner approval before products and prices appear, and location shared at sign-in with the customer’s permission.

  • QR-code POS on one inventory

    A QR code per size, printable labels and whole-pack scanning — on one inventory engine with row-level locks and a full movement ledger, shared by online and in-store sales.

  • AI photo studio & search by photo

    Automatic background removal and HD enhancement on upload, plus search-by-photo powered by vision descriptions and embeddings.

  • Payments & delivery

    Razorpay (UPI, cards, net banking) or cash on delivery; delivery priced by weight, with Delhivery live rates, automatic booking and tracking.

  • GST, Tally & supply chain

    GSTR-3B and GSTR-2B reports, Tally Prime export, WhatsApp low-stock reorders to suppliers, and a commission-based seller programme.

Stack

  • Next.js 15
  • React 19
  • TypeScript
  • Tailwind CSS
  • FastAPI
  • SQLAlchemy 2
  • PostgreSQL
  • Expo / React Native
  • OpenAI API
  • Razorpay
  • Delhivery API
  • WhatsApp Cloud API
  • Docker
  • Playwright
  • pytest

Private client codebase — walkthrough available on request.

Internal tool · Kellton

GenAI Presales Assistant

Proposal drafting from client briefs

An internal Generative AI tool that drafts proposals from client briefs and past engagements — cutting proposal generation time ~60% and reclaiming 15+ hours a week for a presales team of 10+.

faster proposal generation
~60%
reclaimed per week
15+ hrs
  • LangChain
  • GPT-4
  • Python
Production · Kellton

RAG Conversational Agents

Semantic search with an evaluation harness

Conversational agents on OpenAI embeddings and pgvector semantic search, with a retrieval-quality evaluation harness benchmarked against keyword baselines — plus guardrails and latency/cost monitoring in production.

higher retrieval accuracy vs keyword baselines
40%
  • OpenAI embeddings
  • pgvector
  • PostgreSQL
  • Python
  • LLM evaluation
  • Guardrails

05 / Skills

Python-first. LLM-native.

Comfortable in whatever cloud the customer already runs — GCP, Azure or AWS.

  • Python
  • Claude API
  • OpenAI API
  • LangChain
  • RAG
  • pgvector
  • AI Agents
  • Copilot Studio
  • FastAPI
  • Django
  • PostgreSQL
  • Streamlit
  • Docker
  • GitHub Actions
  • GCP
  • Azure
  • AWS
  • AI / LLM

    RAG, agents, evaluation and guardrails — from prototype to production.

    • Claude API (Anthropic) (core)
    • OpenAI API (core)
    • LangChain (core)
    • RAG pipelines (core)
    • AI Agents (core)
    • Copilot Studio
    • Embeddings
    • pgvector
    • Agentic Code Migration
    • Voice AI
    • LLM Evaluation & Guardrails
    • Prompt Engineering
    • NLP
    • Text Classification
    • TensorFlow
    • Scikit-learn
    • NLTK
  • Languages & Backend

    Python services built to stay up — and to be handed over.

    • Python (core)
    • FastAPI (core)
    • SQL
    • C++
    • JavaScript (Basic)
    • Django
    • Flask
    • REST API design
    • OOP
    • System Design
    • pytest / TDD
  • Data & Frontend

    From query tuning to the dashboards stakeholders actually open.

    • PostgreSQL (core)
    • MySQL
    • SQLite
    • Firebase Firestore
    • Pandas
    • NumPy
    • Matplotlib
    • Seaborn
    • Streamlit
    • React.js
    • Jinja2
  • Cloud & DevOps

    Shipping into client tenants within their security and compliance constraints.

    • GCP (GCS, Firebase) (core)
    • Microsoft Azure (core)
    • Docker (core)
    • AWS (EC2, S3)
    • GitHub Actions
    • CI/CD
    • Git
  • Ways of working

    The forward-deployed part of the job.

    • Client requirement discovery
    • Rapid prototyping
    • Stakeholder demos
    • User training and handover
    • Agile / Scrum
    • Production ownership

06 / Research

Published research.

Two national-conference publications in NLP and computer vision.

  1. Published

    National Conference on Advances in Computer Science & Engineering

    Semantic Vector Space Model for Text Classification Using Progressive Learning Network (PLN)

    Multilingual NLP text-classification model using NLTK and the PLN algorithm, outperforming traditional classifiers in accuracy and processing speed.

    • NLP
    • Text classification
    • NLTK
    • Multilingual
  2. Published

    National Conference on Innovation in Engineering & Technology

    Detection of Deepfakes Using MesoNet

    CNN-based deepfake detection system on the MesoNet architecture for images and video; recognised for practical media-integrity applications.

    • Computer vision
    • CNN
    • MesoNet
    • Deepfakes

07 / Education

Education & certifications.

Degree

B.Tech, Computer Science & Engineering

SRM Institute of Science and Technology

Chennai, India

to

Certifications

  • AWS Solutions Architecture Virtual Experience

    AWS

  • Deloitte Technology Virtual Experience

    Deloitte

  • Accenture Strategy Consulting Virtual Experience

    Accenture

  • Python & Django Framework

    Udemy

  • Data Structures & Algorithms

    LetsUpgrade

Portrait of Vansh Singla

08 / Contact

Let’s put your AI idea into production.

Hiring for a forward deployed or applied AI role — or have a GenAI use case that needs to ship? I’d love to hear about it.

Email me
vanshsingla94@gmail.com

Open to Forward Deployed Engineer roles

Based in
Gurgaon, India
Time zone
IST · UTC+5:30
Local time