Forward Deployed Engineer · AI / ML · Senior Software Engineer
I turn ambitious ideas into software that ships.
AI agents, RAG and voice systems, and the products, backends and apps around them.
From “we have an idea” to architecture, engineering and production.
Have an idea, an existing product, or a system that needs fixing? Let’s spend 30 minutes figuring out what should actually be built.
Products I’ve helped build
Real products. Real systems. Real engineering.
- 01AI / agents
- 02RAG
- 03Voice AI
- 04Multi-tenancy
- 05Payments
- 06Ledgers
- 07Real-time systems
- 08Mobile
- 09Workflow engines
- 10Production backends
I don’t just build features.
I build the system around them.
A booking button is easy.
A booking system that handles payments, staff, wallets, timezone rules, cancellations, notifications, mobile apps and reconciliation is different.
That’s the type of work I gravitate toward.
Forward deployed · AI / ML
Forward deployed. AI‑first.
Most AI work breaks in the last mile: the data is messy, the permissions are real, and the workflow belongs to people who never asked for a chatbot.
That’s where I work. I embed with the team that owns the problem — as an engineer, as a consultant, or as the only engineer on the product — and ship AI into the workflow it has to live in.
- 01
Embedded
Inside your team, your repos and your constraints. Full-time, consulting, or as the sole engineer.
- 02
AI in production
Agents, RAG and voice pipelines, some at production scale with high parallel session counts.
- 03
The whole system
The product, backend, data and integrations around the model, not just the prompt.
When software needs to reason.
I’ve spent a significant part of my work building systems around LLMs — not just wrapping an API around a chat box. Agents, RAG, voice pipelines, automation, and enterprise and government AI systems.
The model is one component.
The system around the model is where the engineering lives.
The question is rarely “which model should we call?” It’s what should happen when the model is wrong, slow, uncertain, retried, disconnected or given something it shouldn’t see.
Agentic AI
Agent architectures where models reason through tasks, use tools, maintain state and interact with external systems.
- LangChain
- LangGraph
- Deep Agents
- Tool calling
- State management
- Agent orchestration
- Workflow automation
RAG
Retrieval is not just embeddings. The answer is only as good as the context the model is given.
- Ingest
- Chunk
- Index
- Retrieve
- Context
- Generate
- Evaluate
Voice AI
The complete voice pipeline, including real-time interaction and session management — some of it at production scale, with large volumes of voice interactions and high parallel session counts.
VAD → STT → orchestration → LLM → TTS
- LiveKit
- STT
- TTS
- VAD
Case / 01 — Featured
Glamease
Booking is easy.
Making money, time and people agree is hard.
Case
A salon marketplace and partner platform where customers discover salons, book appointments and pay, while partners manage bookings, staff, billing, earnings and settlements.
Sole developer · Apr–Sep 2026
Built
- Customer marketplace
- Partner console
- Customer Android app
- Partner Android app
- Booking lifecycle
- Wallet
- Staff
- Scheduled jobs
The hard part
- Payment reconciliation
- Idempotent money movement
- Booking state transitions
- Timezone-correct booking logic
- Web + Android from one codebase
More cases
- Case / 02
YumiGlow
“What happens when three businesses need one operational brain?”
Multi-tenant administration + a standalone financial ledger.
React · NestJS · PostgreSQL · Double-entry ledger
- Case / 03
Site Master
“A platform where the customer defines the software.”
Dynamic entities. Forms. Workflows. Approval chains. Reports. Multi-tenancy.
Next.js · NestJS · PostgreSQL · CASL · BullMQ
- Case / 04
PawTales
“The booking model changes when the customer has four legs.”
Pet grooming booking platform.
React · NestJS · PostgreSQL
- Case / 05
Elysian Plaza
“Commerce connected to a wallet.”
Salon supplies ecommerce platform.
React · NestJS · PostgreSQL · Android · Cashfree
- Case / 06
Ashirwad
“A restaurant shouldn’t need a marketplace to run its own orders.”
Food ordering platform with backend, admin and mobile application.
React · React Native · Expo · NestJS · MySQL · Socket.io
- Case / 07
Sampoorna India
“A marketing website with something useful behind it.”
Next.js import-compliance website with an import duty calculator.
Next.js · TypeScript · Tailwind
What people come to me for
People usually come to me with one of these.
“We need AI inside our existing product.”
→ LLM / RAG / agents / automation
“We have an idea, but don’t know how to build it.”
→ Product architecture + MVP
“Our backend is becoming difficult to maintain.”
→ Architecture + backend engineering
“We need someone who can actually own the whole thing.”
→ End-to-end product engineering
“We need a system that can handle real users.”
→ Scalability + infrastructure + production engineering
Something else?
Tell me about the project →
How I work
I work well when the requirements aren’t finished.
Startups rarely hand you a 40-page specification.
Sometimes it’s a Figma file. Sometimes it’s a GitHub repository.
Sometimes it’s a founder saying: “Can we make this work?”
That’s where I tend to be useful.
- 01
Understand
What are we actually trying to change?
- 02
Reduce
What needs to exist for the first useful version?
- 03
Design
Architecture, data, APIs, workflows and failure cases.
- 04
Build
Backend, frontend, AI, integrations and infrastructure.
- 05
Ship
Deployment, observability and the boring details.
- 06
Iterate
Real users reveal the requirements nobody wrote down.
Open source
I don’t only build for clients.
I learn by building things other developers can use. GitHub and npm are part of how I work: packages and projects that grew out of real product problems.
- Package · npm
neo4jd3-graph
The neo4jd3-graph npm package is a tool designed for visualizing Neo4j graph data. This package utilizes the power of D3.js (version 4.2.1) for rendering interactive and dynamic graphs representing relationships within Neo4j databases. The visualization i
- Package · npm
shopping-cart-provider
The shopping-cart-provider component leverages React's Context API to make a configured shopping cart available throughout a React component tree. This component can be imported directly from the shopping-cart-provider
- Repo · buildwithjamu
campaign-web
React + TypeScript + Vite frontend for the outbound campaign platform.
- Repo · buildwithjamu
outbound-campaign-calling-backend
NestJS backend for outbound calling campaigns: scheduling, customer management and call simulation on PostgreSQL + Redis.
A few things I care about
- 01
Don’t put money in a float.
If the system handles money, represent money properly.
- 02
Don’t let the frontend own business rules.
The backend should know what is legal.
- 03
Don’t call something multi-tenant because it has a tenant_id.
Tenant isolation is a system property.
- 04
Don’t use AI for deterministic work.
Let models reason where reasoning helps. Let software enforce what must be correct.
- 05
Don’t build everything before talking to users.
Ship the smallest useful version first.
About
I’m Bala.
Forward Deployed Engineer · AI / ML · Senior Software Engineer
I’ve spent my career moving between product development, backend systems, AI and whatever problem was blocking the team. I’ve built for startups, enterprises and government environments.
Sometimes I write the API. Sometimes I design the architecture.
Sometimes I’m debugging why money disappeared from a wallet.
Sometimes I’m figuring out how an AI agent should behave when the user asks it to do something unexpected.
I like all of it.
Have something difficult to build?
Tell me what you’re working on. It doesn’t need to be fully specified.
30 minutes · No sales pitch · Just figuring out whether I can help.
Prefer email? hello@balareddy.dev