Akshat Agrawal
Verified Expert in Engineering
Artificial Intelligence Engineer and Developer
Bangalore, India
Toptal member since August 14, 2026
At TransFi, Akshat built a conversational-payments AI platform from zero and led a team of 3 engineers. He is a lead AI engineer with over 6 years of experience in back-end engineering, machine learning, and production LLM systems for fintech and healthcare. Akshat's expertise spans Python, AWS, and distributed systems, and he reduced end-to-end response latency by 4.5x while at TransFi.
Portfolio
Experience
- Machine Learning - 6 years
- Django - 5 years
- Natural Language Processing (NLP) - 5 years
- Artificial Intelligence (AI) - 5 years
- Python - 5 years
- Retrieval-augmented Generation (RAG) - 2 years
- FastAPI - 2 years
Preferred Environment
AWS IoT, Docker, OpenTelemetry, Langfuse
The most amazing...
...AI platform I've built is a conversational payments system at TransFi that reduced response latency by 4.5x and supported 1,000+ messages per minute.
Work Experience
Lead AI Engineer
TransFi
- Built TransFi's conversational-payments AI platform from zero and led a 3-engineer team, owning agent orchestration, deterministic payment workflow, observability, latency, and cost optimization.
- Built WhatsApp, Telegram, and web money-transfer agents across seven countries and eight or more languages, orchestrating 14 specialized AI agents over 10 or more payment/account workflows.
- Reduced end-to-end response latency 4.5x (18 seconds → 4 seconds) and reduced LLM cost per conversation by around 35% via single-call routing, model selection, and pipeline parallelization.
- Built a 13-step money-movement pipeline where the LLM handled conversation, extraction, and confirmation while back-end services executed validations, state transitions, rollbacks, and audit trails over authenticated, HMAC-signed internal API calls.
- Built a dependency-graph field model that automatically invalidated downstream payment fields on user edits, enabling safe corrections without manual agent handling.
- Built channel-agnostic ingestion/delivery normalizing WhatsApp Business API, Telegram, and web JSON into one schema over AWS FIFO SQS; load-tested for 1,000+ messages/min and 500+ concurrent conversations with sub-50-millisecond ingestion.
- Implemented per-conversation FIFO ordering, poller/dispatcher backpressure, DynamoDB idempotency (24-hour TTL), and visibility-timeout redelivery.
- Architected a FAQ/RAG platform for 100+ tenants on Weaviate with metadata-filtered isolation, hybrid vector/BM25 retrieval, embedding-provider circuit breakers, and hierarchical tenant/user configuration.
- Added OpenTelemetry traces, Langfuse LLM traces, structured logs with PII redaction, input safety classification, content filtering, audit logs, and an LLM-as-judge eval design.
Senior Software Engineer
Bot MD
- Architected a production multitenant RAG system on tenant-isolated stores with metadata-filtered, multi-step retrieval. Reduced manual support load by approximately 40%.
- Built a team-leader router with Pydantic-validated handoffs to FAQ, scheduling, and data-collection agents. Guarded against invalid LLM routing while preserving multi-turn memory and durable state.
- Built 2-phase calendar scheduling (slot-filling to booking) and document-grounded FAQ flows using long-context grounding and strict source-only prompting.
- Designed a multi-provider abstraction across Bedrock, Claude, and OpenAI with provider selection, secret-redacting telemetry, and config-driven client onboarding.
- Instrumented agent turns with Langfuse and OpenTelemetry, enabling traceability across routing, retrieval, and LLM calls.
Machine Learning Engineer III
Bright
- Built Python/Django back-end services for a US consumer credit card product and rolled it out to an existing user base of approximately 100,000 users. Owned APIs and production workflows across the credit-card lifecycle.
- Built a sub-second in-app underwriting microservice powering credit decisions for AI-based personal loans.
- Mentored junior engineers and led design/code reviews across back-end and ML systems.
- Built a real-time subscription-plan recommendation engine from user financial data, increasing customer acquisition by approximately 10%.
- Built income/expense prediction and analytics services to forecast cashflow patterns and surface savings gaps across the user base.
Software Developer
MathWorks
- Enhanced MathWorks' in-house C++ multithreaded messaging infrastructure to support asynchronous workflows with return values.
- Extended Simulink's Variant Configuration Analysis tool with new Variable Groups functionality.
- Worked in core development as well as customer-facing roles.
Experience
Job Finder
Education
Bachelor and Master of Technology Degree in Information Technology
Indian Institute of Information Technology, Allahabad - Allahabad, India
Skills
Libraries/APIs
Pydantic, Asyncio
Tools
Claude, Amazon Simple Queue Service (SQS)
Languages
Python, Python 3, C++, SQL
Frameworks
Django, LangGraph, LlamaIndex
Platforms
Langfuse, AWS IoT, Docker
Paradigms
REST, Microservices
Storage
Amazon DynamoDB, Elasticsearch, MongoDB, PostgreSQL, Redis
Other
OpenAI, FastAPI, Natural Language Processing (NLP), Retrieval-augmented Generation (RAG), AI Agents, Software Engineering, Artificial Intelligence (AI), Machine Learning, Large Language Models (LLMs), OpenTelemetry, Weaviate, Amazon Bedrock AgentCore, GraphQL APIs, Deep Learning, Prompt Engineering, Recommendation Systems, LangChain, Agno, Pgvector, CI/CD Pipelines
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring