Divyansh Jain
Senior AI Engineer | Building AI Agents that manage money @ Zeni | GenAI, LLMs, RAG, MCP | Python, GCP | ex-Tech Lead (13 members) | 3× 0→1
Bengaluru
6+ years of work experience
May 20242 yrs 4 mosPresent
Pune, Maharashtra, India
Building AI agents and the backend systems that power them at Zeni, the AI bookkeeping platform for startups.
CREDIT ASSESSMENT AGENT — Vertex AI Agent Builder
- Scores every tenant from cash flow, P&L, balance sheet before card approval
- Unlocked 6.7× customer scale (75 → 500+ tenants); ~$80M projected volume, ~$2.4M gross interchange
- Automated ~2,000 hrs/yr of underwriting
- Internal MCP server on Google ADK exposes financial data as agent tools — LRU-cached, feature-gated by tenant; agentic RAG in production
ZENI CARDS — 0→1, Debit & Credit
- First issued cards: issuance, ledger, interchange + cashback engine, spend analytics
- ~$12M annual volume; monthly volume +36% in Q1; 99.9999% uptime, p95 < 200ms
- One card engine across debit and credit (Strategy pattern); webhook processor factory for 6+ Unit (BaaS) event types; multi-tenant namespace isolation
MARK BILL AS PAID — Spend Management
- Saved ~$180K/yr; cut reconciliation 5 hrs/month across 150+ controllers; QBO sync
Tech: Python · GCP · Google ADK · Vertex AI Agent Builder · MCP · LangChain · Unit.co · QuickBooks Online
Apr 20242 mosMay 2024
Bengaluru
Oct 20237 mosApr 2024
Mumbai, Maharashtra, India
Led the GES (Global Expansion Strategy) and PLS (Product Level Strategy) initiatives — GTM tooling for Indian brands entering international markets. Managed 3 FTEs.
- Built semantic retrieval (Sentence Transformers + embedding quantization) matching products to international market categories — foundation of the recommendation pipeline
- Consolidated 8 microservices into one workflow: 10× faster, report creation 12 hrs → 1 hr
- Cut cloud costs 70% (bare-metal migration)
- Multi-modal listing pipeline: LangChain + OpenAI orchestration, Detectron2/YOLO image classification
Jan 20221 yr 5 mosMay 2023
Led 13 engineers across 2 pods. Shipped WhatsApp Flow Builder 0→GA in 3 months; 30+ D2C brands in 6; 100M events/day at 99.97% uptime; AWS cost −40%.
OUTBOUND TEAM - Flow Builder + Conversational AI
- Shipped 2-way conversational AI bots on the RASA framework (Deep Learning + NLU) — a first in the chatbot industry at the time. Automated 90%+ of inbound chat traffic; campaigns ran some of the highest conversion rates in D2C
- Built Flow Builder 0→GA in 3 months — brands create custom interactive flows on WhatsApp and Instagram; lifted customer satisfaction and retention by up to 10%
- Designed LLD / HLD to handle 100M+ events/day
- Event-driven architecture with Factory and Adapter design patterns for maintainability
- Onboarded 30+ D2C brands within 6 months of launch
SERVICES TEAM - Platform Stability, Cost, Observability
- Cut AWS cost by 40% via Kubernetes capacity planning and RDS Aurora IO optimization
- 99.97% uptime across all apps
- Centralized logging on the EFK stack — 10+ applications, designed for 1TB/day ingestion
- ALBs with WAF - secured systems against DDoS and brute-force attacks
- Cut bug resolution time by 70% with Prometheus + Grafana + New Relic monitoring and Slack alerts
- Optimized analytics SQL queries by 50× using query plans, partitioning, and indexes
Team & Process
- Managed two teams via ClickUp; tracked sprint velocity and burndown
- Code quality gates with Pylint + SonarQube + Jenkins (>80% coverage)
- DSMs, weekly planning, retros, monthly 1-1s
Jun 20217 mosDec 2021
Led LimeKit team of 5 people to implement analytics and background jobs
● Designed analytics architecture from scratch, enabling loading rapid insights into all the products
● Introduced RBMQ with Celery for asynchronous processing of weekly analytics
● Worked on MVP of Flow Builder
● Designed a Load testing framework
● Mentored 5 interns
Apr 20213 mosJun 2021
Member of the ACS (Apps Common Services) team, building common features for all the applications
IAM
● Automated keycloak using Ansibles and Helm, enabling the deployment of this service over distributed systems with a single command
● Secured Tableau using SAML, centralising security layer over distributed systems
● Implement WDC for Tableau to secure pushing data
CONFIG SERVER
● Implemented end to end a spring app to save and centralise configurations of distributed systems. Enabled the user to just update config at one place and broadcast over all the systems
Jul 20191 yr 10 mosApr 2021
Member of the RAF(Reflex Analytics Fabric) team, working on building a No code platform for Big Data
CDAP
● Worked on ingestion plugins for Elasticsearch, enabling pushing into indexes, automating index creation over multiple versions of Elasticsearch
● Created a marketplace of all plugins using Python and Shell scripts, automated pushing the required plugins to each client
RAF-POV
● POV of the RAF product, showcasing the TTV of the product
● Created big data pipelines using Hive, Scala, Spark
● Worked with 1.5 TB data per hour (18B rows)
● Designed and optimised Big Data pipelines to process whole Mumbai's network traffic in just an hour and provide detailed insights into it
SECURITY
● Enhance security layer of Elasticsearch to provide RBAC over CRUD operations
● Directly contributed to OpenSource Opendistro framework
● Integrated with LDAP, Kerberos and Ranger
● Deployed it over distributed systems
● Created a detailed manual for clients over how to use the plugins
Divyansh’s AI Stack
Education
Bachelor of Engineering, Computer Engineering
Thapar Institute of Engineering & Technology
20154 yrs2019
General Management Program, Business Administration and Management, General
Stoa
20232023
Join Divyansh and 10k+ other professionals on
Tal is the app where bosses meet engineers & builders directly, without middlemen
Download Tal Boss