Lead AI Engineer Intern @ Capmob

I build agentic AI systems that ship to production.

Sachin Suresh, an AI / ML Engineer working on LLM agents, RAG pipelines and ML-driven decision systems, from the model loop down to the cloud infrastructure that runs it.

Chat with my portfolioRésumé
harness · sample agent trace
CapScore credit-scoring engine at Capmob
50param

CapScore credit-scoring engine at Capmob

gating UnBind in CI
299tests

gating UnBind in CI

in a from-scratch agent runtime
6tools

in a from-scratch agent runtime

B.E. Computer Science, DSCE
8.86CGPA

B.E. Computer Science, DSCE

01 · Experience

Shipping AI where it has to be right: lending, credit and operations.

  1. Lead AI Engineer Intern

    Capmob Financial Services · Hybrid

    Jun 2026 – Present

    • Architected Capmob AI, an SME lending platform on FastAPI + AWS Bedrock: an adaptive interview engine, a deterministic 50-parameter CapScore credit-scoring model with lender-policy fit, and event-driven document ingestion (S3, SQS, Postgres) with concurrent classification and sub-minute extraction.
    • Automated partner onboarding with Zoho: referral CRM lead sync, Zoho Sign agreements with template prefill, and webhook handling with missed-webhook reconciliation and capped retries; added KYC autofill from PDF uploads in a single classify-and-extract call.
    • Managed AWS infrastructure (ECS/Fargate, RDS, ALB, Route 53/ACM, Secrets Manager, KMS, DynamoDB) with CI security gates (Semgrep, Gitleaks, pip-audit, coverage thresholds); built the company website on a custom headless CMS, saving $300+/month in licensing.
    FastAPIAWS BedrockPostgreSQLS3 / SQSECS FargateDynamoDBZohoGitHub Actions

  2. Junior Developer Intern

    XTrail Consulting Services · Remote

    Jan 2026 – Jun 2026

    • Built production dashboards (React + Flask) visualizing operational trends across manufacturing pipelines with sub-second response times.
    • Designed agentic n8n automation workflows integrating custom AI agents, structured JSON parsers and multi-condition decision logic for end-to-end process orchestration.
    ReactFlaskn8nAI Agents

02 · Selected work

Agents, retrieval and the unglamorous parts that make them reliable.

July 202601

Harness

Custom AI agent runtime, built from scratch

A tool-calling agent runtime written from scratch (no LangGraph or AutoGPT) on Groq's gpt-oss-20b, with a dedicated safeguard model and two-tier persistent memory.

  • Tool-calling agent loop wiring 6 tools (web search, weather, calculator, sandboxed code exec, file I/O, email) into a 10-iteration reasoning loop.
  • Guarded by a dedicated gpt-oss-safeguard-20b model that screens user input, every tool call and the final output.
  • Two-tier memory: an LLM-distilled episodic log feeding a 384-dim FAISS semantic index for persistent cross-session recall.
PythonLangChainGroqFAISSHuggingFaceAI Agents
Source
March 202602

UnBind

AI-powered legal contract analyzer

A full-stack contract analyzer that ingests PDF, DOCX and photographed contracts, flags risky clauses, answers questions with verified citations and reminds you before deadlines are missed.

  • Ingests PDF, DOCX and photographed contracts via Groq vision OCR, fanning out clause-by-clause risk analysis in parallel with SSE progress streaming across an 8-stage pipeline.
  • Citation-verified Q&A engine using HyDE retrieval with character-offset source citations, plus a deadline-reminder system with deterministic date parsing across 8 resolution classes and an idempotent GitHub Actions sweep.
  • Payments hardened with HMAC-SHA256 signature verification and atomic quota enforcement, backed by a 299-test suite (281 pytest + 18 vitest) gated in CI by ruff, eslint, tsc and Gitleaks; also shipped as an npm CLI REPL.
Next.js 15FastAPIMongoDBGroqChromaDBRAGRazorpay
Live Source npm

More experiments

Solvent

An AI finance controller built for the Razorpay AI Buildathon (Track 04). It reconciles merchant settlements across Razorpay, ledger and bank data, classifies GST treatment, forecasts settlements and cash position, and answers questions through a retrieval-backed agent.

Python · FastAPI · PostgreSQL · Groq

FundBridge

An AI-powered fintech marketplace connecting businesses seeking loans with banks looking for qualified borrowers, through eligibility scoring, bank matching and funding recommendations.

Next.js · TypeScript · FastAPI · SQLAlchemy

FinSight

An OpenEnv, Gym-style environment for training RL agents on financial document analysis: anomaly detection, KPI extraction and compliance risk assessment.

Python · Reinforcement Learning · OpenEnv · Docker

RagSum

A RAG chatbot assistant paired with a self-built text summarization tool.

Python · LLMs · RAG · FAISS

Chat With Docs

RAG-based application that analyzes uploaded documents and answers queries about them using LLMs.

Python · LLMs · RAG · Streamlit

FairFare

A system designed to reduce ride-booking cancellations during peak hours using ML insights.

Python · Machine Learning · Next.js · TypeScript

Loan Approval Prediction

Predicts loan approval outcomes using supervised ML models.

Python · Scikit-learn · Pandas · Streamlit

House Price Prediction

Machine learning model that predicts house prices from property features.

Python · Scikit-learn · Pandas · Streamlit

Finance Dashboard

Dashboard for tracking and managing personal finances through bank statement analysis.

Python · Pandas · Streamlit · Data Science

Song Recommendation System

Recommends songs based on user-selected tracks using ML techniques.

Python · Machine Learning · Recommendation Systems

03 · Stack

From fine-tuning with QLoRA to deploying on Fargate.

Languages

  • Python
  • TypeScript
  • Go
  • C++

AI / ML

  • PyTorch
  • Scikit-learn
  • NumPy
  • Pandas
  • OpenCV
  • PEFT
  • QLoRA

Agentic & LLM

  • LangChain
  • LangGraph
  • LangFuse
  • HuggingFace
  • Ollama
  • AWS Bedrock
  • n8n

Frameworks

  • FastAPI
  • Pydantic
  • sentence-transformers
  • Weights & Biases
  • Next.js

Data & Vector Stores

  • PostgreSQL
  • MongoDB
  • ChromaDB
  • FAISS
  • Redis

Cloud & MLOps

  • AWS
  • Docker
  • GitHub Actions
  • Vercel
  • MLflow
  • Airflow
  • DagsHub

04 · Recognition & education

Competing, practicing, learning.

  • 2024

    Runner-up, BotCraft

    Runner-up among 100+ teams for a Discord automation bot in Python with an event-driven architecture and slash-command integrations.

  • March 2025

    Shortlisted, Great Bengaluru Hackathon

    Shortlisted while competing among 1000+ teams.

  • 2024

    CodeChef: 100 Days Streak, Peak Rating 1210

    Competitive programming on CodeChef: earned the 100-day coding streak badge with a peak rating of 1210.

  • 2025 – 2026

    LeetCode Daily Challenge Badges

    Consistent DSA and competitive programming practice on LeetCode: 50 & 100 Days coding challenge badges in 2025, and the 50 Days badge in 2026.

Education

Dayananda Sagar College of Engineering

Bachelor of Engineering in Computer Science

8.86

CGPA / 10

Oct 2023 – July 2027
Bangalore, India

05 · Portfolio duel

Think you can out-build me?

Upload your résumé. I’ll pull your GitHub and an AI panel judges us blind, round by round.

youvsme

06 · S.A.C.H.I.N

Don’t skim. Ask.

This portfolio runs its own retrieval-augmented assistant over everything on this page. Ask it what you’d ask me in an interview.

How the assistant works