30×
Faster dashboard delivery
4.5–6 min → 15–30 sec on 300k+ records
AI & Full Stack Engineer / Faridabad, HR, India
AkshatBhatia
I build agentic systems and data infrastructure for industries where getting it wrong isn't an option.
01 / About
About
Full-stack engineer who ended up deep in compliance infrastructure and agentic systems. At OnFinance AI I built the product end-to-end — dashboards, APIs, MongoDB pipelines, and an AI agent layer serving clients like BSE, Navi, and ICICI. I like working on the hard parts: the 4.5-minute query that needs to be 15 seconds, the agent that needs to flag violations before it replies, the MVP that needs to ship in 10 days. That's where I do my best work. Currently looking for what's next — especially teams building at the edge of AI and regulated industries.
90%+
Data extraction accuracy
Up from ~20% with ML-assisted validation
$4.2M
Funding round supported
Pre-Series A from PeakXV Partners
10 days
Compliance platform MVP
Full pipeline: ingestion → evaluation → audit
02 / Experience
Experience
Optimized analytics APIs on 300k+ MongoDB records via multi-stage aggregation pipelines — cutting dashboard response times from 4.5–6 min to 15–30 sec (up to 30× improvement).
Nov 2024 - PresentRemote
Full Stack Engineer
OnFinance AI
- Built the entire frontend from scratch with Next.js, TypeScript, and Tailwind CSS, delivering role-based dashboards and regulatory workflows that enabled compliance teams to process cases 30% faster.
- Optimized analytics APIs on 300k+ MongoDB records via multi-stage aggregation pipelines — cutting dashboard response times from 4.5–6 min to 15–30 sec (up to 30× improvement).
- Migrated AI agent architecture to a Deep Agents framework with skill-based routing, reducing deployment time by 70% and consolidating 47 specialized tools into 8 reusable sub-agents.
Feb 2024 - Jul 2024Remote
Backend Engineering Intern
Wokelo AI
- Built FastAPI + Python pipelines that extracted structured financial data from PDFs using computer-vision and NLP models, enabling automated research document analysis.
- Improved LinkedIn profile validation accuracy from ~20% to 90%+ by developing rule-based and ML-assisted validation logic.
- Fine-tuned Meta LLaMA-3 models for internal AI automation workflows; built DOCX and PPTX report generation pipelines from structured JSON inputs.
ConsultingRemote
AI Systems / Full Stack Consultant
Confidential U.S. Client
- Co-built a production-ready AI agent compliance validation platform MVP in 10 days — covering regulation ingestion, reviewer workflows, real-time violation surfacing, and PDF audit reporting.
- Architected a multi-stage regulation ingestion pipeline: Mistral OCR → section reconstruction → GPT semantic chunking → knowledge-graph triple/ENM extraction → Qdrant vector indexing. Constrained LLM usage to ingestion and report generation only, keeping runtime evaluation deterministic and sub-second.
- Built the FastAPI REST + WebSocket API layer: session management, turn ingestion, real-time flag streaming before each agent reply, and SHA-256 hashed PDF audit reports via WeasyPrint. Containerized the full stack with Docker Compose.
03 / Skills
Skills
Frontend / Backend / Data & AI / Tools
Frontend
ReactNext.jsTypeScriptJavaScriptTailwind CSSHTML/CSSViteTanStack
Backend
FastAPIFlaskNode.jsREST APIsWebSocketMongoDBSQLiteQdrant
Data & AI
LLM Fine-tuningNLPComputer VisionVector SearchKnowledge GraphsPandasNumPy
Tools
GitDockerPrismaJupyterVS CodeCursorClaude CodeLangGraphLangChainRabbitMQ
04 / Projects
Projects
AI-powered vocabulary tracker with spaced repetition and adaptive quiz generation. Replaced a traditional FastAPI + MongoDB + WebSocket stack with Convex's reactive database.
LexiKing
AI-powered vocabulary tracker with spaced repetition and adaptive quiz generation. Replaced a traditional FastAPI + MongoDB + WebSocket stack with Convex's reactive database.
~60% less server-side code
TanStack StartTypeScriptConvexGemini Flash
Consulting Project
AI Compliance Platform
Real-time compliance validation system for AI agent conversations. Detects regulatory violations turn-by-turn and surfaces corrective guidance before the agent's next reply.
Full MVP in 10 days
FastAPIReactTypeScriptQdrantWebSocketDocker
Supabase — Open Source
Contributed logo and dark-mode display fixes across Supabase documentation and the main website.
Shipped to production
Open SourceDocumentation