Aditya Mishra
Sr. Technical Architect & AI Engineer
8+ years building production AI systems end-to-end, from classical ML and NLP to Generative AI, LLM-powered applications, and Agentic AI.
Systems thinker.
AI builder.
I bring both AI fundamentals and the Systems Engineering depth to take models from prototype to production, working across data pipelines, APIs, infrastructure, and model integration trade-offs.
8+
Years of experience
3
Companies shipped at
Experience
Pitchwiz
AI-powered conversational roleplay platform for sales training. Led end-to-end development of proctoring, script reading, multi-language support, and real-time video-based conversational AI.
- ▸Scaled to 100K+ roleplays per month and up to 500 concurrent sessions
- ▸Shaped proctoring and script reading features based on on-ground usage; users were gaming scores without them
- ▸Added multi-language support for non-English-fluent sales reps who struggled with evaluation setup
AI Agent Builder Framework
Internal platform enabling developers, CS, and product teams to build and ship agents without reinventing infrastructure. Built on the insight that agents differ only in tools, system prompts, and context.
- ▸Built agent harness with model fallback, retries, context compaction, rate limiting, memory, and sub-agent composition
- ▸Integrated MCP support with filtered tool exposure per agent to preserve context window and reduce noise
- ▸Added safety middleware covering prompt injection detection, PII redaction, output sanitisation, and platform-specific guardrails
Voice Agents
RAG-based voice agent for sales conversations delivering factually grounded responses and executing user-driven actions in real time.
- ▸Built retrieval pipeline over product knowledge bases with vector search, grounding all responses in client-approved content
- ▸Integrated inline action execution (quote generation, poster sharing) within live conversations
- ▸Designed for real-time latency constraints across STT, LLM, and TTS with full observability
Core Service
In-house authentication and authorisation service in Go, powering every user and every request across the platform.
- ▸Supports phone-OTP, email-password, SSO (SAML and OAuth2.0), and custom login flows with JWT and API key issuance
- ▸Flexible RBAC model with Super Admin, Company Admin, App User, and client-defined custom roles governing access across the platform
- ▸Engineered in Go for sub-millisecond latency, sitting on the critical path of every authenticated request
Real-time Analytics Platform
End-to-end analytics infrastructure built from the ground up, from raw events to actionable insights in under a few minutes.
- ▸Real-time data processing pipelines using Kafka, Flink, and ClickHouse delivering insights within minutes of raw events
- ▸Designed full data sourcing, ETL pipelines, queuing, retries, observability, and scheduling
- ▸Integrated customised Apache Superset dashboards into both the customer app and admin panel
Sales Reels
Short-form video platform solving tacit knowledge sharing in enterprises.
- ▸Built social engagement features (likes, shares, click tracking) and content performance analytics for creators
- ▸Added AI-powered content enhancements including tag generation, summarisation, and title suggestions
- ▸Built admin content moderation and approval workflow with user rewards for approved content
Trust & Data Compliance
PII compliance across multiple microservices to meet enterprise data security requirements.
- ▸Designed custom encryption-at-rest for PII fields across services, protecting data even in the event of a breach
- ▸Introduced field masking alongside encryption to preserve search, filter, and sort on sensitive fields
- ▸Used Unleash feature flags for controlled, reversible rollout across all services
Life Insurance Underwriting Automation
Deep learning system automating underwriting decisions for life insurance proposals, reducing manual intervention across the pipeline.
- ▸Built ETL pipeline for schema transformation from OLTP to analytics-ready format
- ▸Developed deep learning model flagging proposals for approval, rejection, or human review
- ▸Significantly reduced manual intervention and processing time across the underwriting pipeline
Underwriting Automation Module
Tools and dashboards for controlling automation flow and visualising underwriting results for client teams.
- ▸Created early claims detection models integrated into the core decision pipeline
- ▸Designed decision systems combining results from Automation and Claims Detection modules
- ▸Worked directly with clients to integrate the module into their existing systems
Credit Scoring Pipeline
End-to-end credit scoring pipeline for a B2MicroSME lending platform, from raw OLTP data to a deployed scoring API.
- ▸Built OLTP to OLAP ETL with Airflow orchestration and tree-based ensembles (XGBoost, CatBoost)
- ▸Automated digital marketing report generation processing 130GB+ data and producing 34K+ reports in 6 hours
- ▸Deployed as a Flask API with JWT auth and maintained CI/CD via CircleCI with full pytest coverage
ML Across Domains
Applied ML across diverse problem domains for multiple clients.
- ▸Student churn prediction from engagement signals
- ▸PAN card OCR using Mask R-CNN and Tesseract
- ▸Supreme Court judgment annotation tool covering 1950 to 2019
Education
University of Mumbai
B.E in Computer Science
OCPJP Oracle Certified Professional Java SE 6 Programmer
Oracle
Skills
Core Domains
Languages
API Frameworks
AI & ML
Databases
Cloud & Infrastructure
Data & Streaming
Practices
Writing
Boost your Image Classification Model
Techniques for improving deep learning image classifier accuracy: progressive resizing, transfer learning, mixup augmentation, and test-time augmentation.
Metrics to Evaluate Your Machine Learning Algorithm
A practical guide to the most important evaluation metrics for ML models and when to use each one.
Training a Deep Learning Model on Handwritten Characters using Keras
Building a labeled dataset from handwritten character images and training a deep learning model with Keras from scratch.
Get in touch
Have something
to build?
I am open to full-time roles, contracts, and interesting conversations. The best way to reach me is by email.
mishraaditya6991@gmail.com