From the customer's first call —

Bhumil Modi

Forward Deployed AI Engineer

— to agents in production.

Enter ↵
BM
Get in touch

Forward Deployed AI Engineer · Agentic systems

I deploy agentic systems inside the customer.

On the weekly call with your stakeholders, then building the agent pipelines, services and frontends myself — working POC in two weeks, live beta inside five months.

See the casesResume ↓
$ mail bhumilmodi2002@gmail.com

How an engagement runs

How an engagement runs

  1. 01 · Kickoff

    Embed

    Weekly calls with the stakeholders and the product manager. Ambiguity becomes an agent design.

  2. 02 · Week 2

    Prototype

    A working proof of concept on the customer's real inputs.

  3. 03 · Month 2–2.5

    Build

    A functional application: the agent, its guardrails, the product around it.

  4. 04 · Month 4–5

    Beta

    Live with users the customer chooses.

client products
20+
from empty repository to live beta
10+
domains
10
products as project lead
6
commits since Feb 2025
3,792

Approach

How the agent gets built

  • #1 Embed

    Start on the call

    Requirements arrive ambiguous. I sit with the people who own the problem and turn it into an agent design.

  • #2 Design

    Structure first, model second

    Planners, tools and staged pipelines give the agent a shape. The model works inside it, not instead of it.

  • #3 Guard

    Gates before output

    Verification gates, bounded repair and fail-closed steps. A bad draft stops at the gate, not at the user.

  • #4 Measure

    Evals, not vibes

    Deterministic checks plus model judges, run against fixed datasets, so a prompt change is measured rather than eyeballed.

  • #5 Integrate

    Into their stack

    Auth, tenancy, provider abstraction, rate limits and concurrency — what an agent needs to survive a real customer.

  • #6 Ship

    Past the demo

    Beta with real users, release pipelines, and an SDK so other teams can embed the agent.

Work

Agents in the field

A selection of what I've deployed. Clients stay unnamed.

  • 01

    Clinical content agent

    Turns protocol documents into patient-facing scripts, speech and video.

  • 02

    Listing-video planner

    Plans, narrates and renders property videos from listing photos.

  • 03

    Kitchen vision pipeline

    Reads photos of a room and draws plan, elevation and axonometric sheets.

  • 04

    Embeddable avatar assistant

    An AI video-avatar SDK other products drop into their app.

  • 05

    Sales-outreach agents

    Campaigns, lead handling and automated appointment setting.

  • 06

    Meeting intelligence

    Pre-call briefs and live-call analysis.

  • 07

    Grant discovery

    Search, match and apply.

  • 08

    Biopharma agent fleet

    Regulatory gap analysis ahead of a submission.

TheAgentic's own platforms

  • CortexON

    ★ 450+ · open source

    Open-source generalised agent for everyday task automation.

    Frontend contributor

    View on GitHub ↗

  • TheAgenticBench

    ★ 50+ · open source

    Open-source digital-worker framework for agent-driven research and process automation.

    Frontend contributor

    View on GitHub ↗

  • TheAgentic Console

    in-house

    Developer console for TheAgentic's AI infrastructure.

    Built the entire UI

Cases

Four decisions

Four decisions from client work.

  1. IArgus

    clinical · content agent

    Coverage by construction

    Generated content had to reach patients without a reviewer reading every word.

    Case I · Argus · clinical · content agent

    Coverage by construction

    Situation
    Generated content had to reach patients without a reviewer reading every word.
    Decision
    A fact → plan → write → verify pipeline. The planner guarantees each fact is used exactly once; a deterministic gate checks coverage; the model only judges faithfulness.
    Result
    Engineered to need no manual review.
  2. IIAriadne

    generative video · planning agent

    Same brief, same cut

    A model-written video plan could invent assets and quietly drop scenes.

    Case II · Ariadne · generative video · planning agent

    Same brief, same cut

    Situation
    A model-written video plan could invent assets and quietly drop scenes.
    Decision
    A deterministic planner owns the plan; the model may only patch it, and invented references are rejected.
    Result
    The same brief now produces the same video.
  3. IIIDaedalus

    vision · drawing pipeline

    The pipeline I built, then retired

    Turn photographs of a room into dimensioned drawing sheets.

    Case III · Daedalus · vision · drawing pipeline

    The pipeline I built, then retired

    Situation
    Turn photographs of a room into dimensioned drawing sheets.
    Decision
    I built a measured geometry pipeline — then, when image-model rendering beat it, replaced it with a model call per view behind a content-addressed cache.
    Result
    Keep what wins, even when you built the loser.
  4. IVHermes

    agent SDK · release

    An agent other teams can embed

    An AI avatar assistant had to live inside other companies' products.

    Case IV · Hermes · agent SDK · release

    An agent other teams can embed

    Situation
    An AI avatar assistant had to live inside other companies' products.
    Decision
    A framework-free SDK core with a React-free transport, shipped through a gated CI pipeline with dev and stable channels.
    Result
    v1.0 to public release.

Record

Record

  1. May 2025 – nowTheAgenticForward Deployed AI Engineer20+ client products across ten domains; project lead on six.
  2. Nov 2024 – May 2025TheAgenticSoftware EngineerBuilt the entire UI for TheAgentic Console; frontends across seven concurrent engagements.
  3. Feb – Oct 2024ScaleGenAISenior Software EngineerSole owner of the flagship product's frontend architecture.
  4. Oct 2023 – Jan 2024Tally GroupSoftware EngineerBilling-simulation app for Australian utility clients.
  5. Mar – Aug 2023SleevesUpSoftware Engineer InternFull-stack on two products.
  6. 2019 – 2023Vellore Institute of TechnologyB.Tech CSE, AI & MLGPA 9.08 / 10
Agentic
  • Workflow design
  • LLM orchestration
  • Provider abstraction
  • Verification & repair loops
  • Evals
Languages
  • Python
  • TypeScript
  • Go
  • SQL
Backend
  • FastAPI
  • Node.js
Frontend
  • React
  • Next.js
  • Tailwind CSS
Platform
  • PostgreSQL
  • Redis
  • Docker
  • AWS
  • Auth0

Working together

Bring me in on the first call.

Role
Forward Deployed AI Engineer
Based
Ankleshwar, India · IST (UTC+5:30)
Works
Remote, alongside your stakeholders
First version
A working POC within two weeks
Domains
Ten so far, from clinical to CAD
Get in touchResume ↓
$ mail bhumilmodi2002@gmail.com